predictors of depression - university of...

115
2 | Page PREDICTORS OF DEPRESSION A study submitted in partial fulfilment of the requirements for the degree of MSc Data Science at THE UNIVERSITY OF SHEFFIELD by PRANSHU BHASIN September 2016

Upload: buithien

Post on 08-Mar-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

2 | P a g e

PREDICTORS OF DEPRESSION

A study submitted in partial fulfilment

of the requirements for the degree of

MSc Data Science

at

THE UNIVERSITY OF SHEFFIELD

by

PRANSHU BHASIN

September 2016

3 | P a g e

Abstract

Background. A great deal of previous research has highlighted the prevalence of depression

among older adults and its further complications associated with elderly population.

Increasing ageing population has made it even more important to explore predictors of

depression as questions related to the main causes and treatment of depression in older

adults are not completely determined.

Aim. To find predictors of depression within the English Longitudinal Study of Ageing (ELSA)

from the ten factors considered in this study which are as follows. Demographic factors

(age; gender; marital status; children), health factors (insomnia; self-rated health; long

standing illness), other factors (loneliness; financial strain; alcohol consumption).

Design and Methods. By using data from all 6 waves of the English Longitudinal Study of

Ageing, a longitudinal analysis was performed to find predictors of depression by using

descriptive statistics, bivariate and multi variate analysis using SPSS. The predictive ability of

these factors related to depression were further explored by data mining algorithms in

Weka. Different visualisations were produced using SPSS, Excel and Tableau to have a

better understanding about what predicts depression in older adults.

Results. The four predictors out of ten considered factors in this study were: Self-rated

health, loneliness, insomnia and financial strain. Both of the datamining algorithms used in

this study were able to produce good accuracy rates for predicting depression using these

ten factors. Tree visualisation showed that 53 depressed respondents and 2953 non

depressed respondents were present throughout all waves of ELSA. The status of each of

the factor associated with the respondents that were depressed throughout all waves of

ELSA was found to be same at the beginning and at the end of the longitudinal study for all

factors except factors related to alcohol consumption and financial strain.

Conclusion. Among demographic factors, age and gender were insignificant for most of the

waves, whereas marital status was insignificant throughout all waves of ELSA. The factor

related to children was not found to be associated with depression even at bivariate

analysis. Among health factors, insomnia and self-rated health were significantly associated

with depression whereas long standing illness was insignificant throughout all waves. And

among other factors, loneliness and financial strain were significantly associated with

depression whereas alcohol consumption was found to be not significant for each of the

wave of ELSA.

4 | P a g e

Acknowledgements

I would like to thank my supervisor - Professor Peter Bath, as without his help it

would have been very difficult to write this dissertation. He clarified all my doubts

very patiently. I genuinely appreciate his help and support.

Also, no acknowledgement made by me can be complete without mentioning my

heartiest thanks to God and my family for their constant love, support and blessings.

5 | P a g e

Contents Chapter One: Introduction and Context ........................................................................... 8

1.1. Ageing Population ......................................................................................................... 8

1.2. Depression ...................................................................................................................... 8

1.3. Depression in older adults ............................................................................................ 9

1.4. Importance of the study topic .................................................................................... 10

1.5. Research Aims and Objectives ................................................................................... 10

1.6. Structure of the Dissertation Chapters ..................................................................... 11

Chapter Two: Literature Review ..................................................................................... 12

2.1. Considered Risk factors.............................................................................................. 12

a) Gender .............................................................................................................................. 12

b) Age .................................................................................................................................. 13

c) Marital Status ................................................................................................................... 14

d) Children ........................................................................................................................... 15

e) Insomnia........................................................................................................................... 16

f) Self-rated health ............................................................................................................... 16

g) Long standing health problem (Long standing illness/disability/infirmity) .................... 17

h) Loneliness ........................................................................................................................ 18

i) Financial Strain ................................................................................................................. 19

j) Alcohol Consumption ....................................................................................................... 19

Chapter Three: Methodology ........................................................................................... 21

3.1. Quantitative Research Approach .............................................................................. 21

3.2. ELSA Background ...................................................................................................... 21

3.3. Considered Variables .................................................................................................. 21

3.4. Statistical Techniques ................................................................................................. 22

a). Descriptive Statistics ....................................................................................................... 22

b). Bivariate Analysis ........................................................................................................... 23

c). Multivariate Analysis ...................................................................................................... 23

3.5. Data Mining Techniques ............................................................................................ 24

a). Support vector machines ................................................................................................. 25

b). Decision trees ................................................................................................................. 25

3.6. Data for depressed or not depressed throughout all waves (Subset data) ............. 25

3.7. Tree diagram and the status of each factor .............................................................. 26

Chapter Four: Data analysis and results ......................................................................... 27

4.1. Statistical Analysis of ELSA....................................................................................... 27

6 | P a g e

4.2. Prevalence of depression within the ELSA ............................................................... 28

4.3. Prevalence of depression by age and gender within the ELSA ............................... 29

4.4. Univariate Analysis ..................................................................................................... 31

4.5. Prevalence of depression by factors .......................................................................... 34

4.6. Distribution of depression variable by factors broken down by age and gender .. 45

a). Depression by marital status broken down by age and gender ....................................... 45

b). Depression by children broken down by age and gender ............................................... 46

c). Depression by Insomnia broken down by age and gender .............................................. 47

d). Depression by self-rated health broken down by age and gender .................................. 48

e). Depression by Long standing illness broken down by age and gender .......................... 49

f). Depression by Loneliness broken down by age and gender ............................................ 49

g). Depression by Financial strain broken down by age and gender.................................... 50

h). Depression by Alcohol broken down by age and gender ............................................... 51

4.5. Logistic Regression ..................................................................................................... 58

4.6. Comparison of the predictors of depression ............................................................. 67

4.7. Data mining techniques .............................................................................................. 67

4.8. Tree diagram ............................................................................................................... 70

Chapter 5: Discussion ........................................................................................................ 76

5.1. Predictors (The most significant risk factors) .......................................................... 76

a) Self-rated health ............................................................................................................... 76

b) Loneliness ........................................................................................................................ 76

c)Insomnia............................................................................................................................ 77

d)Financial Strain ................................................................................................................. 77

5.2. Insignificant factors - Not at all significant .............................................................. 78

a). Long standing illness/disability/infirmity ....................................................................... 78

b). Marital status .................................................................................................................. 79

5.3. Mostly insignificant factors ........................................................................................ 79

a) Gender .............................................................................................................................. 79

b) Age .................................................................................................................................. 80

c). Alcohol............................................................................................................................ 80

5.4. Not associated factor ................................................................................................... 81

a) Children ........................................................................................................................... 81

Chapter Six: Conclusion and Recommendations ............................................................ 82

6.1. Conclusion .................................................................................................................... 82

6.2. Limitations of this study ............................................................................................. 82

7 | P a g e

a). From the data perspective ............................................................................................... 82

b). From the methods perspective ........................................................................................ 83

c)From the results perspective ............................................................................................. 83

6.3. Strengths of this study ................................................................................................ 84

6.4. Recommendations for Future Research ................................................................... 84

References ........................................................................................................................... 85

Appendix ........................................................................................................................... 105

Forms ................................................................................................................................ 115

8 | P a g e

Chapter One: Introduction and Context

1.1. Ageing Population

In the 20th century, the world has witnessed a remarkable increase in the population

due to several factors such as decreased mortality rates, increased life expectancy,

lower fertility rates, and lesser immigration (Lunenfeld & Stratton, 2013). If the rate

at which the world population is increasing continues through this century, sooner

the 21st century would be called as the ageing century (Christensen, Doblhammer,

Rau, & Vaupel, 2009). In the United Kingdom only, the number of older adults aged

65 and over is now more than the number of children of age less than 15. It is also

estimated that the population of older adults in England only is expected to rise by 39

percent over the coming twenty years (Banerjee, 2014). Progressive ageing

population in the recent decades have given rise to some new challenges in the field

of global public health (Hill, Pérez-stable, Anderson, & Bernard, 2015). It has also

been noted that the prevalence of health problems in the older population has

remarkably increased over the past century which leads to increased healthcare costs

for the society (Jokela, Batty, & Kivimäki, 2013). Among different physical and

mental health problems faced by elderly, depression in older adults, which is

associated with additional ageing problems is considered to be one of the primary

contributors to increased healthcare expenditures and is expected to be one of the

most prominent cause of additional healthcare cost in the developed countries by the

end of 2030 (Maideen, Sidik, Rampal, & Mukhtar, 2015).

1.2. Depression

When a person stops looking out for the meaning and purpose of his/her life and

creates a situation where everything seems meaningless for him/her gives rise to

depressive symptoms (Hodges, 2002). Depression is primarily a psychiatric

condition which significantly affects the mental and physical health of the patient

(Blazer, Hughes, & George, 1987). Depressive symptoms are also the most frequent

psychiatric symptoms and major cause of decreased quality of life (Blazer, 2003). A

combination of different factors of personality and frame of mind can cause

depression in people which can be identified based upon clinical or psychosocial

approach. In the area of mental health, questions related to depression such as how it

9 | P a g e

is classified, what are the root causes and treatment are not fully resolved yet

(Goldstein, & Rosselli, 2016). Although, depression has been explained as an

individual understanding problem which differentiates it from physical healthcare

(Reitzes, Mutran, & Fernandez, 1996) however, it is not only a personal health

failure but a failure in the way health care field and society have understood and

responded to the problems related to this field (Lewis, 1995). The prevalence of

depression across cultures is further evident from various studies and surveys of

different countries performed by WHO (Offici, 2001).

1.3. Depression in older adults

From the classical times depression has been one of the most common mental

disorder in later life, affecting up to 15% older adults of age 65 and above

(Livingston, Hawkins, Graham, &Blizard, 1990). All over the world depression in

elderly has become a more serious health problem due to three main reasons: first,

the ageing population and eventually increased number of older adults with

depression, second, its consequences are recognised by WHO on the “global burden

of disease”(Lopez & Murray, 1998), and third, achievements in the field of

neuroscience have further increased knowledge and understanding about serious

consequences and sufferings that accompany depression in old age (Blazer, Hughes,

& George, 1987). Depression in old age is considered to be more harmful because

neither the patient nor the clinician in most of the cases is able to identify its

symptoms in presence of other health problems (Loder, 2009). Depression when co-

exists with other health problem in elderly leads to several other complications and

further deteriorate the condition of existing health problems, for instance, it can lead

to functional disability when it co-exists with physical health illness (Gureje,

Ademola, & Olley, 2008). It has also been noted that the response of elderly to the

medical treatment is less and depression happens to be more severe if the first onset

of depression is in old age in comparison to older adults that suffers from recurrent

depression in old age (Blazer, Hughes, & George, 1987). Depression in older adults

is also identified as an independent predictive factor for suicide and in the suicide

cases of older adults aged 75 and over, 80% of them had depressive symptoms

(Alexopoulos, 2005).

10 | P a g e

1.4. Importance of the study topic

Depression is not a part of normal ageing but a condition that can be prevented (Scott,

1989). Depression is considered to be one of those mental health problems that can

be treated effectively and efficiently if diagnosed and treated on time without a much

delay (Bower, 1986). There is a need to fill the gap that exists between depression

recognition and its intervention in older adults as the diagnosis of depression in older

people is difficult (Murray et al., 2006). Depression in older adults is further

complicated by the fact that majority of older people do not report their depressive

symptoms by considering them as a part of ageing instead of considering them as

sign of a mental health problem that needs to be considered and treated (Vanessa et

al., 2006) due to which depression in older adults for most of the cases are under-

diagnosed and under-treated (Rosenvinge, 1988). Apart from this, few studies have

suggested that older people intentionally do not report or admit these symptoms due

to the stigma attached with the mental disorders (Blazer, 2003) which again results in

often under-recognised and under-treated disorder (Anderson, 2001) even when the

fact is that major cases of depression are found in older adults. Also, for clinicians, to

predict prognosis of depression is poor when it comes to old age (Subramaniam &

Mitchell, 2005). Therefore, deep understanding and knowledge about predictors of

depression will help both clinicians and patients to have a better understanding about

this mental disorder and to be aware about its predictors for recognising depression at

early stages and where possible trying to work out on the factors that can prevent

depression.

1.5. Research Aims and Objectives

The main aim of this dissertation is to find predictors of depression in older adults.

The aim of the dissertation is met by achieving a number of objectives as follows.

First, to review the previous literature related to depression and its association with

different factors in order to select the most relevant factors discussed with depression

and could be the potential predictors of depression. Second, to explore the prevalence

of overall depression for all waves of ELSA and with each factor considered in this

study. Third, analysing ELSA dataset for all six waves to find relationship between

depression and chosen interested variables using bivariate analysis and then to

11 | P a g e

identify the major correlates of depression and classifying predictors based on

multivariate analysis using SPSS. Fourth, to identify the individuals who were either

depressed or not depressed throughout all six waves of ELSA to identify predictors

of depression for these individuals to allow a comparison of these results obtained

with predictors of depression identified from the analysis of general population

sample for all 6 waves of ELSA. Fifth, to follow the individuals that were depressed

throughout all waves to determine whether the status of each factor for most of the

individuals were same or have changed through these 12 years of depression. Sixth,

to investigate how accurate results were produced by these factors in predicting

depression using data mining techniques and then to compare it with the accuracy

rates produced by the statistical technique used in this study to identify predictive

ability of these factors. Seventh, to compare the results obtained from this study with

previously published literature. And finally, to identify strengths and limitations of

the present study and to offer recommendations for future research in identifying

predictors of depression.

1.6. Structure of the Dissertation Chapters

This dissertation is structured into six chapters. Chapter One aims to place the

context and significance of the research topic. Chapter two examines different earlier

contributions in this area of research and presents the key points obtained from the

literature for the factors related to depression considered in this study. Chapter Three

offers a description about the data and variables used in the analysis and provides

detailed description of different methods used in this study. Chapter Four presents

various results obtained from the longitudinal analysis of ELSA data. Chapter Five

discusses the key findings obtained from analysing the ELSA data and relate them

with available published research. Finally, Chapter Six offers the conclusion drawn

from this research and identifies the strengths and limitations of the present study to

offer recommendations for future research in this study topic.

12 | P a g e

Chapter Two: Literature Review

There is a large volume of published studies describing the factors related to

depression in older adults. In this section, the key points are highlighted from the

most relevant published literature that relates depression with various risk factors

included in this study.

The literature added in this study has been synthesized from the searches made on

Medline, PubMed, Web of Science and Google Scholar. However, a lot of relevant

literature was also obtained from the “Starplus”: The University’s digital library,

which further provided links to several other resources. The key terms used were

“depression”, “predictors of depression”, “elderly”, “depression in older adults” to

identify the factors which were most widely discussed with depression in previous

published studies. After that, different combinations of search terms were used to

synthesize the most relevant literature for specific ten factors that were identified.

For example, “gender AND depression”; “gender differences in depression”; “gender

AND (old* OR elderly OR aged) AND depression”. Similarly, different

combinations were used for each of the factor.

A number of other factors related to depression were also identified after a review of

great deal of previous research that has focussed on depression and its associated

factors. However, given the aims of the study where depression was to be compared

with each of the factor in statistical as well further analysis, considering more than

ten factors was out of scope for this study. The main other factors identified were:

“having diabetes” (Jacobson, 1993; Williams, 2006); “early retirement” (Karpansalo,

2005; Schofield et al., 2011); “less social participation” (Holtfreter, Reisig, &

Turanovic, 2015; Bourassa, Memel, Woolverton, & Sbarra, 2015), “Lower life

satisfaction” (Lee, 2014; Farakhan, Lubin, & O’connor, 1984); and “no friends”

(Potts, 1997; Seeman, 2000).

2.1. Considered Risk factors

a) Gender

Women are found to be more depressed than men and major cases of depression are

higher in women (Weissman & Klerman, 1985). The number of cases reported for

13 | P a g e

depression in females is almost double than males (Culbertson, 1997). A great deal

of previous research has focused on female gender as a key risk factor linked to

depression in old age (Blazer, Burchett, Service, & George, 1999; Kessler, 2003). A

number of authors have reported that severe cases of depression in women over 65

years of age were higher compared to men of the same age group (Katsumata et al.,

2005; Regan, Kearney, Savva, Cronin, & Kenny, 2013; Maguen, Luxton, Skopp, &

Madden, 2012). A good example of this is the cross country analysis performed by

Velde, Bracken, and Levecque (2010. However, Faravelli, Scarpato, Castellini, and

Sauro (2013) claimed that it was only prior to menopause when women were more

depressed than men. Conversely, Cairney and Wade (2002) argued that prevalence of

depression in females even after the menopause is high compared to males. In their

detailed study of association of depression with gender, Ochoa et al. (1992), reported

that female gender was a significant risk factor only when depression and anxiety

were considered together, and was not a significant factor for depression alone. This

view is supported by Joiner and Blalock (1995) who argued that prevalence of

clinically significant depression is almost same in males and females but symptoms

related to depression and anxiety were found to be more in females. There are, in

contrast, few studies that have reported that gender is not a significant risk factor for

depression; it is just that women with depression suffer more with feelings related to

worthlessness (Dessoki, Moussa, & Nasr, 2011).

b) Age

Numerous studies have attempted to explain how depression is associated with age.

Research performed by Stordal, Mykletun and Dahl (2003) reported that even after

controlling different variables in their study, increase in age was significantly

associated with increased depression. However, an opposite view was provided by

Jorm (2000) where he claims that with ageing the risk of depression decreases as the

person develops more control over his/her emotions in addition to decrease in

emotional response accompanied with normal ageing. This view is supported by

Henderson, Jorm, Korten, Jacomb, and Christensen (1998) in their study where they

concluded that with age depression decreased for each older adult for both males and

females. However, there is an inconsistency in the published results related to age

and depression, for instance, study by Zarit, Gatz and Johansson (1999) claimed that

14 | P a g e

in older adults the tendency to experience depressive symptoms are more common

than actually having clinically diagnosed depression. This inconsistency in the results

is further evident from a study performed by Kim, Shin, Yoon and Stewart (2002),

where they compared the associated factors of depression in rural and urban areas of

Korea, and concluded that age was a significantly associated with depression for

older adults living in urban areas but was not significant for older adults that were

depressed in rural areas. Another good example for this is a research presented by

Danesh and Landeen (2007) on lifetime and one-year depression for all age groups,

where the authors found that both lifetime and one-year depression had the highest

rates for age group (20-24) and (75+) but the lowest rates of depression were found

in the age group (75+) when both types of depression were together taken into

account, although, the highest rates were still for the age group (20-24).

c) Marital Status

The protective effects of marriage play a significant role in preventing depression

irrespective of the gender (Kim & Mckenry, 2002) which is evident from the

literature that has emphasized on the “married” marital status to be inversely

associated with depression (Koenig, 1988; Comstock & Helsing, 1977). However,

existing research also identifies the effects of the same marital status to be different

for both males and females (Harlow, 1991). One the one hand, it is published that

married as the marital status was associated with less depression for both men and

women equally (Stack & Eshleman, 1998). On the other hand, few studies have

reported that married women were less depressed than married men (Glenn, 1975).

And in old age, marital status as “divorced” was most associated with depression for

men, whereas for women it was “widowed” marital status that was most associated

with depression (Kamiya, Doyle, Henretta, & Timonen, 2013). Conversely, Etaugh

and Malstrom (1981) concluded that “widowed” marital status was less associated

with depression compared to “divorced” marital status for women. The effect of

marital disruption was also studied by Bruce and Kim (2002) where they reported

that men were at higher risk of depression than women after a divorce. In addition, it

has also been published that positive interaction and a good quality relationship

being shared in a marriage is equally important (Santini, Koyanagi, Tyrovolas, &

Haro, 2015). In this regard, a study performed by Kronmuller et al. (2011) concluded

15 | P a g e

that marital quality and depression were significantly associated with each other

(Kronmüller et al., 2011). However, Kim and Mckenry (2002) suggested that among

marital quality and marital status, marital status is more significantly associated with

depression.

d) Children

Sociologists have always stressed about the support which older parents can receive

from their children and how having them in the social network may prevent mental

health problem such as depression (Evenson et al., 2005). In this regard, study

performed by Chou and Chi (2004) demonstrated that childlessness was significantly

associated with depression (Chou & Chi, 2004). This theory has been verified by

several other studies, for instance, Sener (2011) demonstrated that depression and

children were associated negatively with each other such that older adults having

children were found to be less depressed and vice versa. Similarly, a study by Guo

(2014) also provides evidence that having more than one child was associated with

lower depression among elderly. In the same vein, Oxman, Berkman, Kasl, Freeman,

and Barrett (1992) from their study suggested that depression was even associated

with the number of children that visited their older parents. Moreover, the attitude

towards childlessness is also associated with depression, COX (2002) illustrates this

point very clearly that negative attitude had more impact on depression for females

than for males. Differences in the gender towards childlessness and its association

with depression was further demonstrated by Connidis and McMullin (1993) where

they showed that females not having children due to circumstances were more

depressed compared to females that have children whereas for males, it was just that

they were less happy compared to males that have children and were not found to be

depressed. Another different viewpoint was offered by Sener (2011) where he

reported that emotional support that older adults receive from their children is

associated with less depression only till the point where exchange of emotional

support is same from both sides of the relationship that is shared among children and

parents.

16 | P a g e

e) Insomnia

A large and growing body of literature has investigated how insomnia and tendency

to be depressed are related to each other (All et al., 2000; Orhan et al,.2012). A very

complex relation exists between depression and insomnia (Gambhir, Chakrabarti,

Sharma, & Saran, 2014). On the one hand, it has been published that insomnia is a

significant risk factor for development of depression and, therefore, treatment of

insomnia at early stages reduces the risk of development of depression (Roberts,

Shema, Kaplan, & Strawbridge, 2000). On the other hand, it has also been reported

that insomnia is a secondary symptom associated with depression, thus, treatment of

insomnia doesn’t have impact on the risk of the development of depression (Ford,

1989). A number of authors have found a significant association between insomnia

and depression (Taylor, Lichstein, Durrence, Reidel, & Bush, 2005; Morin &

Gramling, 1989). A Recent study by Chang et al. (2014) provided evidence that even

“perceived sleep quality” is also associated with depression in elderly, as older adults

with poor perceived sleep quality had higher levels of depression and perceived sleep

quality was an independent predictor of depression in elderly. A study by Pallesen et

al. (2002) compared different measures of physical and mental health among older

adults based on sleep quality and they found that older adults with sleep problems

had much higher number psychological problems and depressive symptoms

compared to those who had normal sleep. In contrary, the study performed by

Neckelmann, Mykletunand & Dahl (2007) reported results from two different health

surveys which demonstrated insomnia to be significantly associated with depression

in only one of the health survey, not in both of them.

f) Self-rated health

Based on self-rated health clinicians have been able to make prognosis of depression,

and in identifying patients who are at the maximum risk of facing long-term

depression (Livingston et al., 1990). Poor self-rated health has been identified as a

key risk factor for depression by different authors (Dendukuri & Cole, 2001; Jae &

Sook, 2006). Moreover, from the results of the study performed by Murrell,

Himmelfarb and Wright (1983), self-rated health variable had the strongest

association with depression in comparison to any other variable used in their study.

Other than self-rated health as a predictor of depression, few studies have even

17 | P a g e

published about depression as a predictor of poor self-rated health by older adults, a

good example of this is the study carried out by Wagner and Short (2014), where

they showed depression as an independent key risk factor for poor self-perceived

health. Additionally, it has been noted that with treatment of depression, the self-

rated health of many patients also improve simultaneously even in the absence of any

physical health improvement (Han, 2002b). Very interesting results were reported by

Ambresin, Chondros, Dowrick, Herrman, and Gunn (2014) that even in the presence

of the current depression variable in multivariate analysis along with demographic

factors, poor self-rated health remained as an independent predictor for future major

depression, and elderly people who indicated fair or poor self-rated health had almost

double risk of major depressive syndrome up to next five years (Badawi et al., 2013).

Current literature has also identified a bi-directional relationship between self-rated

health and depression, such that depression at baseline wave was an independent

predictor of poor self-rated health at follow up waves and poor self-rated health at

baseline was a predictor of depression in follow up waves (Han, 2002).

g) Long standing health problem (Long standing illness/disability/infirmity)

More than one third of disabilities arise from mental health disorders (Druss et al.,

2008). In this regard, research performed by Kivela and Pahkala (2001) to

understand effect of depression on disability, they concluded that depression at wave

1 was not associated with disability in the follow-up waves, however, a new episode

of depression at follow-up wave for someone who was not depressed at wave1 was

associated with disability. The course of depression is found to be more long-term

when it co-exists with a long standing illness (van den Brink et al., 2002). However,

it has also been published that long standing illness causes depression among people

from all age groups (Aneshensel, Frerichs, & Huba, 1984). Existing research

recognizes the prevalence of depression more among people with long standing

illness or which in medical terms is known as chronic illness (Benton, Staab, &

Evans, 2007; Nikolic, 2015). This is evident from the study by Haseen & Prasartkul

(2011) where they reported that people with infirmity and disability had the highest

risk of depression. Individuals with long standing health problems have significantly

greater risk of depression but a complication arises from the fact that in presence of

these physical health problems, depression often remains undetected, and when

18 | P a g e

detected the care and support required from both physical as well as mental health

care is often not coordinated (Hawkes, 2012). Rifel, ävab, Pavlič,, King, and

Nazareth (2010) concluded that over period of 6 months patients suffering from long

standing illness had four times higher risk of depression compared to other

patients. Similarly, a longitudinal study of major depression by Patten (2001) reports

that long standing illness was significantly associated with major depression and risk

of major depression is almost double in the patients with long standing illness.

h) Loneliness

Depression and various other mental health problems can arise from experiencing

loneliness (Adams, Sanders, & Auth, 2004). For instance, a recent study by Bekhet

Zauszniewski (2012) compared the results of overall health of the patients that felt

lonely and not lonely, in which it was found that not much differences existed in the

physical health but significant differences existed in terms of mental health of lonely

and not lonely patients. There is a large volume of published studies describing the

relation between depression and loneliness perceived among older adults (Alpass &

Neville, 2016; Green et al., 1992; Jongenelis, Pot, Eisses, & Beekman, 2004).

Depressive symptoms are more prominent in older adults who are lonely and they

suffer more than adults who are depressed but not lonely (Liu, Gou & Zuo, 2016).

This is further supported by Jaremka et al. (2013) who claimed that lonelier patients

were more depressed. In the presence of loneliness, prognosis for depression

becomes much more difficult (Holvast et al., 2015). Further, depression has also

been noted to be an independent contributor for both emotional as well as social

loneliness (Drageset, Espehaug, & Kirkevold, 2012). Also, Stek et al. (2015) claimed

that chances of depression to be fatal were comparatively more among the elderly

who felt lonely. One the one hand, a longitudinal study by Houtjes et al (2014)

demonstrates that people who had minor level of depression or experienced recent

symptoms of depression eventually became lonely over time. On the other hand,

comparison of two longitudinal studies by Povoski et al. (2013) reports that lonely

people eventually became depressed over time. Similar findings were replicated by

Aylaz et al. (2012) who confirmed loneliness to be an independent key risk factor for

depression. Likewise, the study by Aylaz, Aktürk, Erci, Öztürk, and Aslan (2012)

confirmed that elderly people who felt lonely were eventually more depressed.

19 | P a g e

i) Financial Strain

A great deal of previous research has investigated how financial hardship is

associated with mental health problems and psychological disorders that arise due to

constant stress, hopelessness, and feeling of uncertainty caused by shortage of money

(Pudrovska, 2005; (Mirowsky & Ross, 1999). Old people having financial strain are

found to be more depressed (Lue, Chen, & Wu, 2010). This view is supported by

Chou and Chi (2005) who reported that financial strain was significantly associated

with depression for all older age groups. In an analysis of stressful life events

experienced by elderly, Fiske, Gatz, and Pedersen (2003) found that financial stress

was among the most validated one. Similarly, in a longitudinal study, Lue, Chen and

Wu (2010) found that perceived financial stress was a key risk factor for depression.

In the same vein, a recent study by Hsieh (2015) reported that high economic status

was consistently associated with lower level of depression among older adults. Based

on gender, different studies have published different results, for instance, a

longitudinal study performed by Mandes De Leon, Rapp and Kasl (1994) reported

that financial strain was significantly associated with depression for men only and

found that men with financial strain became depressed over 3 years. However, in

another study by Lue, Chen, and Wu (2010), females were found to be more

depressed who had higher financial strain

j) Alcohol Consumption

Risk of appearance of depressive symptoms is five times higher in older adults who

indulge in heavy amount of alcohol consumption (Saunders et al., 1991). Moreover,

problems related to drinking make older adults more susceptible to other psychiatric

problems such as depression (Johnson, 2000). A study by Bekaroglu,

Uluutku ,Tanriover and Kirpinar (1991) reported that for older adults in Turkey, high

alcohol consumption was significantly associated with higher levels of depression.

Similarly, Graham and Schmidt (1997) reported that alcohol consumption was a

significant risk factor for depression. However, Graham and Schmidt revealed that

only higher volume of alcohol consumption was associated with depression whereas

frequency of alcohol consumption was not associated with depression in older adults.

In contrast to Graham and Schmidt, the authors Bulloch, Lavorato, Williams, and

Patten (2012) argued that any drinking more than moderate level drinking as per

20 | P a g e

standard guidelines were not associated with higher depression, whereas, higher

levels of depression were found among people who were dependent on alcohol

within moderate levels. Conversely, Lang1, Wallace, Huppert, and Melzer (2007)

published that less depression was found for older adults that drink moderate levels

of alcohol. Few researchers have confirmed a significant association between alcohol

consumption and depression to be true only for women (Aihara, Minai, Aoyama, &

Shimanouchi, 2010), however, few researchers have claimed it to be only associated

with men (Bulloch, Lavorato, Williams, & Patten, 2012). Another research carried

by Damian et al. (2012) reported that alcohol consumption and depression were

negatively associated with each other, such that people who drank more were less

depressed. However, few studies have also demonstrated that no significant

association exist between alcohol consumption and depression (Fishleder, Schonfeld,

Corvin, Tyler, & VandeWeerd, 2015)

21 | P a g e

Chapter Three: Methodology

This section provides an overview and justification of the methods adopted in this

study to serve the purpose of the study to find the predictors of depression.

3.1. Quantitative Research Approach

As aim of this study was to find relationship of different factors with depression and

predictors of depression, which involved statistical analysis of ELSA data which was

numeric in nature, therefore a quantitative approach was suitable to achieve the aims

and objectives of this study (D.Scott, 2007).

3.2. ELSA Background

In this study, data from the English Longitudinal Study of Ageing (ELSA) survey

were analysed (Broudeur, Hurrell, Stepinska, Fluffy, & Houxou, 2014). Six waves

of data are available in ELSA – data were collected from the interviews that were

held every two years. The data for the six waves were collected over eleven years -

2002, 2004, 2006, 2008, 2010, and 2012. It is a study of older adults aged 50 and

over, designed to get an insight of older population of England and is an on-going

study related to ageing (Hamer, Batty, & Kivimaki, 2012; Lang et al., 2009). To

collect objective as well as subjective data, questions were grouped into thirteen

modules and a self-completion questionnaires form. Efforts were made to ensure that

data of each wave consistently reflects all age group people in it and data were

refreshed at Wave 3, 4 and 6, therefore not all respondents were present from the first

wave of ELSA. Although, at certain waves new additional modules were added,

however the main purpose of the ELSA survey was to identify changes that occur

over time for the measured variables to better understand ageing in older people of

England (Marmot, 2003). However, it was also designed in such a way that it could

be compared with other longitudinal studies related to ageing in older adults across

the globe (Steptoe, Breeze, Banks, & Nazroo, 2013).

3.3. Considered Variables

Based upon literature discussed, variables related to each of the factor were identified

such that they were present throughout all waves of ELSA. The actual names of the

22 | P a g e

variables considered were, “INDSEX” for (Gender), “INDAGER” for (Age),

“DIMAR” for (Marital Status), “SCCHD” for (Children) “PSCEDC” for (Insomnia),

“HEHELF” for (Self-rated health), “HEILL” for (Long standing

illness/disability/infirmity), “PSCEDE” for (Loneliness), “SCQOLI” for (Financial

Strain), and “HEALA” (Wave 1) and “SCAKO” (Wave 2 to 6) for (Alcohol) and

“PSCEDA” for (Depression). The question asked and the response options

(categories) for each of the variable at the time of interview can be seen in Appendix

A

3.4. Statistical Techniques

Throughout this thesis, data analysis step was first performed for wave 1 and then for

all other waves and subset data (obtained for people who were depressed or not

depressed throughout all waves). Due to word limit consideration, detailed

description and interpretation for the results of wave 1 has been described and only

summarised comparative results were stated for all remaining waves and for subset

data. Also throughout this thesis yellow colour has been used to represent non

depressed respondents and grey colour has been used to represent depressed people

in all visualisations as it has been published in the literature that depressed people are

most attracted to grey colour and yellow is most associated with not depressed

people (Carruthers, Morris, Tarrier, & Whorwell, 2010)

a). Descriptive Statistics

Descriptive statistics is used to understand the characteristics of data in a much better

way from which many ideas and assumptions can be made by just having a glimpse

at these statistics. Descriptive statistics make use of various graphical techniques

(Kadane, 2016)

In this study, as all variables used were categorical, frequency statistics was used a

lot to summarize data about each variable. Descriptive statistics for all of the

variables was undertaken and prevalence of depression throughout all waves of

ELSA was explored. The data was split by depression variable to further explore the

prevalence of depression based on each of the factor considered in this study. For

wave 1 only, an additional analysis was performed where data were split by age and

23 | P a g e

gender to explore the influences age and gender have on depression associated with

each of the factor considered. Different forms of graphical representations such as

clustered bar charts, simple bar charts and line charts were used to describe data as

well as to visualize the results obtained.

b). Bivariate Analysis

To achieve the aims and objectives of this study it was very important to know

whether any association exists between depression and other factors considered. As

all of the variables in this study were categorical, hence chi-squared test was used to

perform bi-variate analysis.

“Chi-squared test is used to examine independence across two categorical variables

or to assess how well a sample fits the distribution of a known population” (Franke,

Ho, & Christie, 2012, p.449)

The basis of the Chi-squared test is to either accept or reject the null hypothesis. The

Null hypothesis is formulated in such a way that it states that no significant

association exists between categorical variables, and, the aim of the researcher is to

reject this null hypothesis based on the results of the test applied (Vitral, Campos, &

Fraga, 2013).

Chi-squared test was used to identify which of the ten variables had significant

association with depression. In case of each variable with which bi-variable

comparison was made, a null hypothesis was stated as: No association exists between

that variable and depression

c). Multivariate Analysis

Different techniques are available to perform multivariate analysis based upon the

type of data stored in interested variables. In this study, dependent variable

(depression) was categorical; hence logistic regression could be performed (Foster et

al., 2016). Logistic regression will help to identify predictors of depression as values

of dependent variable can be predicted from other independent variables by using

this technique. Also, as depression variable divides all the respondents into two

24 | P a g e

cases- depressed or not depressed, hence logistic regression was applied to identify

which associated variables predict depression in older adults.

Three models were used for logistic regression at each wave of ELSA. The first

model included only demographic factors (Gender, Age, Marital Status and

Children). The second model introduced health related factors (Insomnia, Self-rated

health and Long standing illness) into model 1 to look for whether these could

explain the identified associations at model1. Finally, the third model introduced

other factors (Loneliness, Financial Strain and Alcohol) into model 2 to further

establish which of the factors independently predict depression.

3.5. Data Mining Techniques

Two data mining algorithms namely Support Vector Machines and Decision Trees

were considered in this study. Although any one of the algorithm could have served

the purpose to identify how predictive the factors considered in this study were,

however, to achieve the aims and objectives of this study as precisely and clearly as

possible, two algorithms were used as both of them could produce different accuracy

rates. Overall accuracy, sensitivity, specificity, positive predictive value, negative

predictive value of each algorithm was obtained from the confusion matrix to

investigate how accurate were the factors considered in this study to predict

depression. How they were calculated have been described in the figure below.

25 | P a g e

a). Support vector machines

This algorithm of classification constructs a hyperplane which acts as a boundary to

separate two classes. The distance between this boundary and the data points on each

side of this boundary are called support vectors, from which this algorithm got its

name as Support Vector machine (Yu, Liu, Valdez, Gwinn, & Khoury, 2010).

This study has included SVM algorithm for finding predictors of depression, because,

it is now widely used in disease detection and is a high performance classification

algorithm especially in the field of bioinformatics (Son, Kim, Kim, Choi, & Lee,

2010). In addition, its approach for prediction is very different from logistic

regression. As SVM is not based on probability of classes within the data set as it

manages to find a separation boundary between variables which is opposite to

logistic regression which constructs a regression line and tries to minimize the error

and have a more probabilistic approach (Verplancke et al., 2008)

b). Decision trees

Decision trees divide the dataset into nodes and branches (Coussement, Bossche, &

Bock, 2014). Due to their hierarchal structure, they make the data set more

interpretable and visually more understandable when compared to other data mining

algorithms. Another reason for including this classification algorithm for this study is

because much has been written about the decision trees efficiency to model partitions

of the datasets which is difficult to achieve through logistic regression and other

prediction algorithm such as SVM (Kingsford & Salzberg, 2008). Decision trees

have a further option of pruning. With help of pruning the size of decision trees is

reduced to only the most relevant section of the tree which has got the maximum

predictive value and is easy to visualise (Quinlan, 1999). In this study, both pruned

and un-pruned decision trees would be considered.

3.6. Data for depressed or not depressed throughout all waves (Subset data)

Data for respondents who were either depressed or not depressed throughout all 6

waves of ELSA were obtained and the same analysis as done for all other 6 waves

was performed on this data and predictors of depression were identified and were

compared with the results obtained for predictors of depression from all other 6

26 | P a g e

waves to check whether the predictors from general sample and for those who were

in same state of depression throughout all waves were same or different, to have a

better understanding about predictors of depression. Data for this purpose were

obtained by merging data files for all remaining waves to wave 1 based on the

variable “idauniq” which was unique individual serial number to uniquely identify

respondents throughout all waves. After this, another variable was computed that

stored value 0 or 1 for the respondents who were either depressed or not depressed

throughout. IF statement was further used to decide the value 0 or 1 for this variable

based on the condition such that if depression variable was 0 for all the waves then it

was given a value of 0 i.e. not depressed throughout and was given a value of 1 when

depression variable was 1 for all waves of ELSA. This data is further referred as

“subset data” throughout the thesis.

3.7. Tree diagram and the status of each factor

From data obtained for the respondents who were depressed or not depressed

throughout, a further analysis was conducted by splitting the data by depression

variable from the first wave and calculating the frequency of depression for second

wave and this was repeated for all other waves till all the observations were made

and based on this tree diagrams were obtained for the respondents who were

depressed or not depressed throughout all waves of ELSA and how respondents

changed their depression status throughout all waves of ELSA.

Further, data were filtered for those respondents who were depressed throughout all

waves of ELSA and status of each of the factor considered in this study was obtained

based on the mode value i.e. the category for each of the factor which had the highest

number of these identified depressed people. This was done to explore and have

better understanding about status of each variable found for those people who were

depressed throughout these 11 years and further to identify if there was any change

in the status of each factor at the beginning nd at the end of this longitudinal analysis

which would further throw light on the predictors of depression.

27 | P a g e

Chapter Four: Data analysis and results

4.1. Statistical Analysis of ELSA

Before beginning with the analysis of depression and other variables, it was

important to know how many respondents were present at each wave of ELSA.

Figure 2 (left) above shows the number of respondents that participated in each of

the wave. Wave 1 had the maximum number of respondents (12099) whereas the

wave 2 had the minimum number of respondents (9432). After this, it was important

to explore how many among these respondents at each wave answered the question

asked for measuring depression. The question asked was “much of the time during

the past week, you felt depressed”. Line chart in the figure 2 (right) above shows the

total number of respondents that answered the question. Again the number (11717)

was found to be the maximum at wave 1. From both of the charts, it can be seen that

the response rate for depression was higher for wave 1 compared to all other waves.

Therefore, in this study wave 1 was used as a base for all the analysis performed.

28 | P a g e

4.2. Prevalence of depression within the ELSA

Figure 3 (top) compares the number of older adults that were depressed and not

depressed at each of the wave. As can be seen from the figure that the highest

number of depressed people (2084) were present in the ELSA at wave 1 and the

lowest number of depressed people were present at wave 6. Similarly, the number of

not depressed people was also highest (9633) for wave 1, and lowest (7704) for wave

2. As mentioned that not all of the respondents answered the question asked for

depression. Therefore, the figure 3 (bottom) below shows the valid percentage of

depressed and not depressed people for each wave. It can be seen that the proportion

of depressed people was found to highest for wave 1 and the proportion of not

depressed people was found to be highest for wave 6.

29 | P a g e

4.3. Prevalence of depression by age and gender within the ELSA

The charts above were produced to explore the prevalence of depression among older

adults based on age and gender. It can be seen from the charts that throughout all

waves of ELSA across all age groups a higher proportion of females were depressed

than males and, the age group 80+ had a higher proportion of both depressed males

and females compared to all other age groups. The differences in the proportion of

depressed and not depressed people were most evident for wave 1.

30 | P a g e

31 | P a g e

4.4. Univariate Analysis

Frequency tables were produced to explore the distribution of the ELSA data for

each of the variable considered in this study. The table 1 below shows the summary

of descriptive statistics carried out for wave 1. It can be seen that more number of

“females” (6764) were present compared to “males” (5335). A higher proportion of

respondents (37.1%) were from age group “50-59” compared to only 11 % of the

respondents from the age group “80+”. Over half of the respondents (56.3%) marital

status was “married”, whereas the category “single” accounts for only 5.6 % of the

respondents. Most of the respondents (86.6%) indicated that “they have children”.

Just over half of the respondents (59.2%) reported their sleep as “normal”. A higher

proportion of respondents (30%) reported their self-rated health as “good” compared

to only 9.2 % of the respondents who indicated their health as “poor”. 55.9 % of the

respondents indicated the presence of “long standing illness”. Just 13.2% of the

respondents felt “lonely”. The proportion of respondents (14.6%) “often” having

financial strain was lower than all other categories of financial strain, and

“sometimes” having financial strain was reported by higher proportion of the

respondents (33.7%). The proportion of the respondents (27.9%) that had alcohol on

daily basis and those (30.4%) who had once or twice a week was not much different.

And the respondents (10.6%) who had alcohol once or twice a month were even less

than the respondents (11.7%) who never had alcohol.

32 | P a g e

33 | P a g e

Similarly, univariate analysis was also calculated for each of the factor for the

remaining waves and for subset data to identify which categories had higher

proportion of respondents which has been summarised in the tables 2 and 3. Similar

pattern of data distribution (“females” (gender); “married”(marital status); “have

children”(children); “good” (self-rated health); “normal”(insomnia); “have long

standing illness”(long standing illness) ; “not lonely”(loneliness)) was found for all

of the remaining waves and subset data for each of the variable other than for the

factors financial strain and alcohol consumption where no consistent category was

identified that had more number of individuals across all waves and subset data.

Hence, the categories with a higher proportion of individuals throughout all waves

and subset data were similar to wave 1.

f m f m f m f m f m

Gender Male 4125 4295 4925 4569 4744

Female 5307 5476 6125 5705 5857

Age 50-59 2925 3434 3428 2560 2937

60-69 2920 2706 3696 3766 3847

70-79 2203 2069 2478 2533 2547

80+ 1123 1134 1147 1060 1063

Marital Status Single 496 574 675 596 670

Married 5233 5345 6077 5704 5862

Remarried 1036 1196 1433 1224 1276

Separated 997 1083 1231 1158 1242

Widowed 1669 1572 1632 1586 1548

Children Have Children 7137 7060 7954 7698 7704

Don't have Children 1034 1091 1258 1224 1259

Insomnia Restless Sleep 3885 3891 3697 3980 3428

Normal Sleep 5328 5569 6833 5668 6491

Self-Rated Health Excellent 1179 2465 1353 1180 1212

Very Good 2598 4072 3091 2902 2892

Good 2934 2331 3375 3094 3156

Fair 1877 529 1994 1792 1919

Poor 706 137 781 757 800

Long standing health Yes 5323 5291 5984 5615 5780

No 4103 4473 5057 4645 4814

Loneliness Lonely 1275 1252 1340 1253 1146

Not Lonely 7941 8207 9188 8397 8768

Financial Strain Often 955 1038 1255 1137 1284

Sometimes 2295 2356 2751 2668 2698

Not often 2187 2420 2836 2664 2592

Never 2665 2315 2336 2447 2292

Alcohol consumption Almost daily 2888 2818 3346 3137 3127

Once or twice a week 2094 2019 2259 2095 2058

Once or twice a month 983 946 1031 1032 1018

Special occasion only 1296 1255 1433 1493 1450

Not at all 883 879 1034 1105 1177

7

687

1735

1771

311

237 549

14

624

1358

0

207

3

1638

682

622

9

522

1872

1947

0

355

6

1352

626

1412

7

312

1642

1854

0

301

2

1838

520

456138

6

216

1330

1288

0

428

1

1620

Variable

0

261

1

1261

219

Each Variable Statistics - Wave 2 to 6 (Frequency - f and total missing values -m)

Wave 2 Wave 3 Wave 4 Wave 5 Wave 6

34 | P a g e

4.5. Prevalence of depression by factors

To explore the prevalence of depression based on different factors considered in this

study, data were split by depression variable and descriptive statistics was calculated

for each of the variable which has been summarised in the table 4 below for wave 1.

35 | P a g e

The table 4 shows that there were more depressed “females” (19.8%) than “males”

(15.3%). The age group “80+” had the highest proportion of depressed respondents

(22.6%) whereas “60-69” had the lowest proportion of depressed respondents

(15.9%). The proportion of depressed people for the first three categories of marital

status were comparable, “depressed-single” (18.1%), “depressed-married (15.0%),

“depressed-remarried” (16.3%) and was almost same for the rest two categories:

“depressed-legally separated” (24.2%); “depressed-widowed” (24.1%). 17.1 % of the

respondents were depressed who fall in the category “have children” compared to

36 | P a g e

15.4 % depressed respondents who don’t have children. 30.9 % of the respondents

were depressed who suffered from “restless sleep” than just 8.7 % of depressed

people with normal sleep. The highest proportion of depressed people (47.5%) was

for the respondents who indicated “poor” self-rated health after “fair” (26.8%). Only

7.8 % of the respondents were depressed who indicated ‘excellent” self-rated health.

Respondents that indicated the presence of long standing illness were more depressed

(23%) compared to depressed respondents (11.2%) without long standing illness. For

loneliness, a much higher proportion of lonely depressed respondents (51.8%) were

there compared to depressed not lonely respondents (12. 4%).The proportion of

depressed people (31.7%) was maximum for the “often” financial strain category

after “sometimes” (16.8%). The proportion of depressed people for alcohol

consumption was higher for respondents who never had alcohol (28.0%) compared

to 14.6 % of depressed respondents who had alcohol “almost daily”.

Additionally, clustered bar charts added below were also produced to better

understand depression with each of the factor and to look for any association that

may depict by looking at them before the chi-square tests were performed to actually

identify any association.

37 | P a g e

38 | P a g e

Similarly, for rest of the waves and for subset data, the prevalence of depression

based on the factors was calculated. The tables 5 and 6 below summarise the results

obtained for remaining waves and subset data. It was observed that the prevalence of

depression based on each factor for all of the remaining waves and subset data was

similar to wave 1 - the proportion of depressed people was higher for “females”

(gender), “80+” (age), “married” (marital status), “have children” (children),

“restless sleep” (insomnia), “poor” or “fair” (self-rated health), “have long standing

illness” (long standing illness), “lonely” (loneliness), “often” or “sometimes”

(financial strain), “not at all” (alcohol consumption).

39 | P a g e

40 | P a g e

Subset data

Further bar charts were created for subset data also to look whether the charts depict

any association.

41 | P a g e

42 | P a g e

43 | P a g e

The bar chart below was produced to look for how depression variable was present in

all different waves to have clear picture about number of depressed people present

throughout all waves in each factor and its category.

44 | P a g e

45 | P a g e

Additional analysis just for wave 1 was carried out further to explore the distribution

of depressed respondents present in ELSA by each factor broken down by age and

gender to look for how depression variable was distributed among each factor based

on age and gender. These charts were created to just to explore the number of

respondents present for each category in the ELSA, they don’t deal with prevalence

or the proportion of depressed people as prevalence of depression has already been

discussed.

4.6. Distribution of depression variable by factors broken down by age and

gender

a). Depression by marital status broken down by age and gender

It can be seen from the figure 8 above that there were higher number of both males

and females present in the ELSA with marital status as “married”, and that too from

46 | P a g e

the same age group “50-59”, less number of depressed people with age group 80+

were present (although the total number of respondents present in the 80+ group

were also lower, that is the reason it has been mentioned above that these charts

doesn’t deal with proportion or prevalence of depression, on the other hand these

were created to have an understanding that how many depressed people were

available for each category based on age and gender in ELSA). Depressed females

were more than depressed males for each age group category other than 80+. It can

also be seen that for legally separated respondents, number of not depressed females

were also higher than males for each age group category.

b). Depression by children broken down by age and gender

It is apparent from the figure 9 that for both gender and for each age group depressed

respondents with children were higher in ELSA population. Only for age group “50-

59” and “60-69” the number of depressed males that “don’t have children” were

47 | P a g e

higher than females. For all other categories more depressed females were present in

the ELSA.

c). Depression by Insomnia broken down by age and gender

It is clearly visible from the figure 10 below that for each age group and for both

genders higher number of depressed people were present who had “restless sleep”.

Depressed females that indicated restless sleep were higher in number for age group

"50-59” compared to all other age groups. For only age group “60-69” the number of

depressed males and depressed females that reported normal sleep were almost same,

rest for each category, depressed females that reported restless sleep were higher than

males.

48 | P a g e

It is clearly visible from the figure that for each age group and for both genders

higher number of depressed people were present who had “restless sleep”. Depressed

females that indicated restless sleep were higher in number for age group "50-59”

compared to all other age groups. For only age group “60-69” the number of

depressed males and depressed females that reported normal sleep were almost same,

rest for each category, depressed females that reported restless sleep were higher than

males.

d). Depression by self-rated health broken down by age and gender

As the figure 11 below shows that the number of depressed females that reported

their health as “fair” was consistently higher for each age group. Also, it shows that

the number for depressed males that reported their health as “poor” and “fair” was

same. For each category number of females that were depressed were higher than

males except for self-rated category “excellent” and “very good” for age group “70-

79”.

49 | P a g e

e). Depression by Long standing illness broken down by age and gender

It is evident from the figure 12 that the number of depressed respondents was higher

for the category “have children” for both genders and for all age groups. A higher

number of depressed and not depressed females were present in age group “50-59”.

In females, age group “70-79” had the second highest number of depressed females

after age group “50-59”, whereas in case of males it was “60-69” age group that had

second highest number of depressed males present.

Figure 12: Long standing illness

f). Depression by Loneliness broken down by age and gender

From the figure 13 below, it can be seen that for each age group, for both males and

females there were higher number of depressed people in the ELSA who were “not

lonely” other than age group “80+” in which depressed people that felt lonely and

50 | P a g e

not lonely were comparable. The highest number of depressed females that were “not

lonely” was from age group “50-59” whereas for males it was “60-69”. Depressed

females that felt lonely were higher in age group “70-79” than all other age groups

whereas depressed males that felt lonely were higher in age group “50-59”.

g). Depression by Financial strain broken down by age and gender

As can be seen from the figure 14 below that the number of depressed respondents

with financial strain as “often” was higher in number for both males and females. For

each financial strain category and age groups, the number of depressed females was

higher than males, except for age group “60-69”and “70-79” where financial strain

51 | P a g e

was “often” and for “70-79” with financial strain as “not often”. Also, it is clear that

for financial categories “often” and “sometimes” there were highest number of

depressed respondents for both genders and for each age group.

h). Depression by Alcohol broken down by age and gender

It can be seen from the figure 15 below that depressed males are consistently higher

than females who drank “almost daily” except for the age group “80+”. Even for the

alcohol consumption category “once or twice a week” there were more depressed

males than females for all age groups other than “50-59”. For rest of the categories,

52 | P a g e

the number of depressed females was higher than males. A higher number of

depressed males were present for the alcohol category “once or twice a week” for

each age group, whereas for females there was no particular category that had higher

number for all age groups.

53 | P a g e

4.4. Chi- squared test: To further identify whether any association exists between

these factors and depression, the chi-squared test was performed between each of the

factor and depression, results of which have been summarised in the table 7 below.

Variable

χ2

df

p

1. Gender 39.617 1 0.000

2. Age 36.903 1 0.000

3. Marital Status 123.856 4 0.000

4. Children 2.518 1 0.121

5. Insomnia 954.304 1 0.000

6. Self-Rated Health 475.957 4 0.000

7. Long standing

illness/disability/infirmity

275.812 1 0.000

8. Loneliness 1456.592 1 0.000

9. Financial Strain 239.152 1 0.000

10. Alcohol Consumption 158.282 1 0.000

54 | P a g e

From the results summarised in the table 7 above it can be seen that there was a

significant association between gender and depression, of the 5149 males, 786

reported having depression (15.3%) compared with 1298 depressed women out of

the 6568 (19.8%) in the study (χ2= 39.617; df=1; p<0.001). A significant association

was also found between age and depression, the depression rate was 16.3% for age

group “50-59”, compared to 15.9% for age group “60-69”, 20.8 % for age group “70-

79” and 22.6 % for age group “80+” in this study (χ2= 36.903; df=1; p= p<0.001).

Similarly, a significant association existed between marital status and depression, the

depression rate was 18.1 % for “single” compared to 15.0 % for “married”, 16.3%

for “remarried”, 24.2 % for “separated” and 24.1 % for “widowed” in the study

(χ2=123.856; df= 4 p<0.001). No significant association was found between children

and depression, of the 9324 adults who had children, 1590 were depressed (17.1%)

compared with 222 out of 1444 adults who don’t have children (15.4%) in the study

(χ2=2.518; df=1; p=0.121). Another significant association was found between

insomnia and depression, of the 4776 respondents who reported restless sleep, 1478

were depressed (30.9%) compared with 605 out of 6938 respondents who reported

normal sleep (8.7%) in this study (χ2=954.304; df= 1; p<0.001).Likewise, a

significant association existed between self-rated health and depression, depression

rate was 7.8 % for the category “excellent”, 8.9 % for “very good”, 14.5 % for

“good”, 26.8 % for “fair”, and 47.5 % for “poor” in the present study (χ2=475.957;

df= 1; p<0.001). Long standing illness also had a significant association with

depression, of the 6515 respondents who indicated that they have long standing

illness, 1501 were depressed (23%) compared with 583 depressed respondents out of

the 5200 who reported they don’t have long standing illness (11.2%) in this study

(χ2=954.304; df= 1; p<0.001). Similarly, loneliness was also found to be

significantly associated with depression, of the 1593 lonely people, 825 were

depressed (51.8%) compared with 1257 depressed respondents out of 10119

respondents who were not lonely (12.4%) in this study (χ2=1456.592; df= 1;

p<0.001). Another significant association was found between financial strain and

depression, depression rate was found to be 31.7% for the category “often”, 16.8 %

for “sometimes”, 12.9 % for “not often”, 11.8 % for “never” in the present study

(χ2=239.152; df= 1; p<0.001). And, a significant association was also found between

alcohol consumption and depression, depression rate was found to be 14.6% for the

category “almost daily”, 14.3% for “once or twice a week”, 22.3 % for “special

55 | P a g e

occasion only”, and 28.0 % for “not at all” in the present study (χ2=158.282; df= 1;

p<0.001).

Similarly, chi-squared test was performed for all other remaining waves which has

been summarised in the table 8 below. Results obtained were consistent throughout

all waves. Each of the factor was found to associated with p-value of less than 0.001

except the factor related to children (p= 0.600 (wave-2); p=1.000 (wave-3); p=0.249

(wave-4); 0.515 (wave-5); p= 0.351 (wave 6), which was found to be not associated

with depression for all six waves of ELSA. However, from the results of chi-squared

test for subset data which can be seen in the table 9 below, it was found that no

association existed between age and depression (p=0.506), also no association was

found between gender and depression (p=0.233) in addition to the factor related to

children (p=0.795) which was found to be not associated in other 6 waves also.

Variable

Wave

2

Wave

3

Wave

4

Wave

5

Wave

6

1. Gender χ

2

df

p

46.803

1

0.000

47.919

1

0.000

43.813

1

0.000

27.730

1

0.000

18.929

1

0.000

2. Age χ

2

df

p

37.721

1

0.000

19.539

1

0.000

21.348

1

0.000

6.873

1

0.009

1.743

1

0.187

3. Marital Status χ

2

df

p

145.291

4

0.000

163.371

4

0.000

121.884

4

0.000

162.000

4

0.000

119.420

4

0.000

56 | P a g e

4. Children χ

2

df

p

0.275

1

0.600

0.000

1

1.000

1.331

1

0.249

0.424

1

0.515

0.871

1

0.351

5. Insomnia χ

2

df

p

592.906

1

0.000

674.563

1

0.000

860.466

1

0.000

632.806

1

0.000

724.517

1

0.000

6. Self-Rated Health χ

2

df

p

724.017

1

0.000

816.298

1

0.000

852.656

1

0.000

778.523

1

0.000

756.811

1

0.000

7. Long standing illness

/disability/infirmity

χ

2

df

p

191.866

1

0.000

214.299

1

0.000

200.477

1

0.000

214.120

1

0.000

241.333

1

0.000

8. Loneliness χ

2

df

p

1345.68

4

1

0.000

1536.73

1

1

0.000

1455.87

4

1

0.000

1489.72

8

1

0.000

1454.53

8

1

0.000

9. Financial Strain χ

2

df

p

171.709

1

0.000

195.358

1

0.000

269.135

1

0.000

268.791

1

0.000

201.250

1

0.000

10. Alcohol

Consumption

χ

2

df

p

82.194

1

0.000

138.185

1

0.000

122.859

1

0.000

142.313

1

0.000

110.395

1

0.000

57 | P a g e

Variable

χ2

df

p

1. Gender 1.420 1 0.233

2. Age 0.442 1 0.506

3. Marital Status 22.756 1 0.000

4. Children 0.067 1 0.795

5. Insomnia 84.485 1 0.000

6. Self-Rated Health 132.688 1 0.000

7. Long standing

illness/disability/infirmity

30.130 1 0.000

8. Loneliness Fisher exact test

was used

Fisher exact test

was used

0.000

9. Financial Strain 47.582 1 0.000

10. Alcohol Consumption 12.928 1 0.000

58 | P a g e

4.5. Logistic Regression

To determine which of the associated factors obtained from the chi-squared test were

most significant and independently predict depression, Binary logistic regression was

carried out to identify predictors of depression. The variable “children” was not

associated with depression from the results of chi-squared test and but here in table it

was added just to have an overall view of each variable although it has to be ignored

for multi-variate analysis.

59 | P a g e

Table above shows that only 2 % of the variation in depression could be explained by

model 1(R²=.020).

In model 1(only demographic factors), each of the demographic factor was

significant: gender (p< 0.001); age (p< 0.001) and marital status (p< 0.001). It

showed that males had 79.3 % of the risk of depression compared to females

(p<0.001; OR=.793; 95% Cl=.712 to .885). For marital status, the only category

which was found to be associated was “married” (p<0.001) with 68.4 % of the risk of

depression compared to “widowed” category (p<0.001; OR=.684; 95% Cl=.586

to .798).

Age group categories “50-59” and “60-69” were significant with (p=.029) and

(p=.012). Age group “50-59” had 80.2 % of the risk of depression compared to “80+”

age group (p<0.029; OR=.802; 95% Cl=.658 to .978) whereas age group “60-69” had

77.7% of the risk of depression compared to “80+” age group (p<0.012; OR=.777;

95% Cl=.637 to .946).

Model 2 (demographic plus health factors) was able to explain 20.8% of the variation

in depression (R²=.208). After adding health factors, only marital status remained

significant (p=.003) whereas both age and gender became insignificant.

Among health factors self-rated health (p<0.001) and insomnia (p<0.001) were found

to be highly significant. Longstanding illness was found to be not significant

(p=.272). Insomniac individuals had 39.06 % higher risk of depression compared to

individuals who indicated normal sleep (p<0.001; OR=3.906; 95% Cl=3.285 to

4.644). Each category of self-rated health was highly significant with depression

p<0.001 and had a lower risk of depression compared to individuals that reported

self-rated health as poor as OR value for each of them was less than 1.

Model 3 (demographic plus health plus other factors) was able to explain 27.4 % of

the variation in depression (R²=.208). Variables that were significant in the presence

of all factors considered were insomnia (p<0.001), self-rated health (p<0.001),

loneliness (p<0.001), and financial strain (p<0.001). This model showed that the

individuals with “restless sleep” had 34.8% higher risk of depression compared to the

individuals with “normal sleep”. Self-rated health was significantly associated with

60 | P a g e

depression (p<0.001). Even, all the categories of self-rated health were also

significant with p<0.001 for each of the category (p<0.001; OR=3.489; 95%

Cl=2.909 to 4.185). Lonely people were 4.44 times more likely to be depressed than

not lonely people (p<0.001; OR=4.449; 95% Cl=3.562 to 5.555). For individuals

with financial strain as “often” (p<0.001) and “sometimes” (p=0.016) there was 21.3 %

(p<0.001; OR=2.133; 95% Cl=1.625 to 2.799) and 13.4% of higher risk of

depression (p<0.016; OR=1.340; 95% Cl=1.055 to 1.701) compared to individuals

who never had financial strain.

For self-rated health each of the category was significantly associated with

depression: ‘excellent” (p<0.001) with 23.3 % of depression risk (OR=.233; 95%

Cl=.154 to .353), “very good” (p<0.001) with 23.9% of depression risk (OR=.239;

95% Cl=.172 to .332), “good” (p<0.001) with 32.2 % of depression risk (OR=.322;

95% Cl=.243 to .427), “fair” (p<0.001) with 51.65 of depression risk (OR=.516; 95%

Cl=.393 to .679) compared to individuals with poor self-rated health.

Similarly, binary logistic regression was carried out for all other remaining waves

and for the subset data, the results were consistent for most the variables. Overall,

from the final model, self-rated health, loneliness, insomnia and financial strain were

significantly associated with depression throughout all 6 waves and in subset data.

Whereas, gender and age had inconsistent results as gender came out to be

significant for wave 3 (p=0.052) and wave 4 (p=0.014) whereas age came out to be

significant only at wave 4 (p=0.016). Long standing illness and marital status were

insignificant throughout all waves of ELSA and in subset data. Alcohol consumption

was also insignificant for each wave of ELSA and in subset data except at wave 6

(p=0.033).

Results obtained for each remaining wave can be seen in the tables below. Results

were summarised in the tables below.

61 | P a g e

62 | P a g e

63 | P a g e

64 | P a g e

65 | P a g e

66 | P a g e

Subset data

Age, gender and children were not added as they were not associated as interpreted

from the chi-squared results

67 | P a g e

4.6. Comparison of the predictors of depression

The figure 16 below shows the results of predictors obtained from wave 1, rest of the

waves and from the subset data. A check sign represents that factor was significant in

that model. A cross represents insignificance and red colour box represents

significantly associated factor with depression in the final model. The figure shows

that insomnia, self-rated health, loneliness and financial strain were consistently

significantly associated with depression in all waves and in subset data.

4.7. Data mining techniques

The classifier SMO which train a support vector classifier in Weka was built by

percentage split of 80 (training) is to 20 (testing) as enough data was available.

Therefore, 80 % of the data was sufficient for the classifier to learn and make

predictions for the 20% data that was used by Weka for testing. From the 20 % data

that was used for predictions, the results produced by SVM were as follows. Out of

1180 instances, 961 were correctly classified whereas 219 instances were incorrectly

classified by the SVM classifier. As clear from the confusion matrix below that

classifier was able to predict 878 instances correctly and 88 instances incorrectly for

the class “not depressed” whereas, the classifier was able to predict only 83 instances

correctly compared to 131 incorrectly classified instances for the class “depressed”.

68 | P a g e

Overall accuracy rate produced by SVM was 81.44 %, the outputs from Weka for

each of the algorithm can be seen in appendix B.

Similarly, the decision tree algorithm J48 available in Weka was used to further

check what accuracy rate could be produced by it using the factors considered in this

study. With pruning, the classifier was able to produce an accuracy rate of 84.661 %.

Out of 1180 instances, the decision tree classifier (with pruning) was able to classify

999 instances correctly and 181 instances incorrectly. The classifier was able to

predict 929 instances correctly and only 37 instances incorrectly for the class “not

depressed” whereas 144 instances correctly and 70 instances incorrectly for the class

“depressed”.

From the figure 17 below, it can be seen that J48 pruned tree showed that loneliness

was one of the most important predictive factor among all other variables as it was

shown as the root node. Another key predictor that can be interpreted from the tree

obtained is insomnia. The tree shows that considering only “not lonely” reached leaf

node of predicting depression as not depressed for approximately 10123 instances

out of which around 1258 instances were misclassified by the classifier. There were

999 instances that could be reached with just considering “lonely” and “restless sleep”

to predict depression, for which 371 were incorrectly classified using this classifier.

Similarly, 594 instances could reach to not depressed class only by considering “not

lonely and “normal sleep”.

69 | P a g e

J48 (without pruning) produced an accuracy rate of 82.711%. Out of 1180 instances,

976 were correctly classified and 204 were incorrectly classified by the classifier. As

apparent from the confusion matrix in the figure 18 below that the classifier was able

to predict 906 instances correctly and 60 incorrectly for “not depressed” class and

144 correctly and 70 incorrectly for class “depressed”.

Overall, decision tree algorithm with pruning option produced the highest accuracy

rate, from which it can be concluded that using data mining algorithms to predict

depression based on the factors considered in this study can overall produce a

maximum of 84.66% accuracy rate.

Confusion matrix obtained for each of the data mining algorithm from the Weka can

be seen the figure 18 below

SVM J48 (pruned tree) J48 (un- pruned tree)

And confusion matrix obtained for logistic regression through SPSS can be seen in

the figure 19 below

70 | P a g e

Further, a comparison was made between the results obtained by data mining

algorithms and statistical technique (which in this study was logistic regression) to

identify which of the techniques have produced better model for predicting

depression based on the factors considered in this study.

On comparison, it was found that the best accuracy rate (85.61%) was produced by

logistic regression. However, its specificity (the ability of the test to correctly predict

people who were depressed) was worse than all other data mining algorithms (just

23.8%) Although, its ability to classify not depressed people was better than all other

algorithms (specificity= 0.973 i.e. 97. 3%). The positive predictive value (correct

prediction of depressed people) was found to best for decision tree -unpruned.

Whereas, negative predictive value (correct prediction of not depressed people) was

found to be best for logistic regression.

4.8. Tree diagram

Tree diagrams in the figure 21 and 22 below shows that 53 respondents (2.54 %) out

of 2084 depressed respondents at wave 1 were depressed throughout all waves of

ELSA and 2593 respondents (26.91%) out of 9633 were not depressed from wave 1

to wave 6.

71 | P a g e

72 | P a g e

73 | P a g e

Status of each factor

Further, the status of each of the variable was identified based on mode value for

wave 1 and wave 6. It can be seen from the bar charts in the figure 23 below that

status for financial strain was changed from “sometimes” to “often”. Another change

in status was for alcohol consumption that changed from “once or twice a week” to

“not at all”. Rest for each of the variable the status was found to be same at wave 1

and wave 6. The associated status for these 53 depressed respondents for all other

variables were: “female” (gender); “married” (marital status); “have children”

(children); “restless sleep” (insomnia); “poor” (self-rated health); “have long

standing illness” (long standing illness); and “lonely” (loneliness). It was not

calculated for the age variable as data were for people who were depressed

throughout all 6 waves and hence increase in age was expected so the status.

74 | P a g e

75 | P a g e

Based on this, the figure 24 below has summarised what was the status of each factor

associated with most of the respondents that remained depressed for these 11 years of

study. In other words, pathway to depression within the ELSA

76 | P a g e

Chapter 5: Discussion

In this chapter the findings about predictors of depression from the longitudinal

analysis of ELSA are discussed and compared with previous published literature.

5.1. Predictors (The most significant risk factors)

a) Self-rated health

One of the most significantly associated factor in this study was self-rated health.

Each of the category of self-rated health was also significantly associated almost

across all waves of ELSA with (p<0.001). This finding was very much expected as

considerable amount of literature has been published on how self-rated health is

associated with depression (Lochen & Rasmussen, 1996; Chow & Chan, 2010). This

finding of self-rated health evidently support the results published by Badawi et al.,

(2013) that reported poor self-rated health at baseline wave was an independent

predictor of major depression at follow up. Similarly, Damián, Barriuso, and Gama

(2008) reported that poor or fair self-rated health was a significant risk factor for

depression in older people. Likewise, Ruo et al. (2006) reported from their study

based on older women, that majority of women facing depression rated their health

as poor or fair. Relationship between self-rated health and depression analysed by

Han (2012), identified a bidirectional relationship between self-rated health and

depression, such that depression at baseline wave was an independent predictor of

poor self-rated health at follow up waves and poor self-rated health at baseline was a

predictor of depression in follow up waves.

b) Loneliness

Another predictor of depression and highly significant factor (p<0.001 for each wave)

associated with depression throughout all waves of ELSA was loneliness. This was

an obvious finding as a large and growing body of literature research has highlighted

the strong association between depression and loneliness (Wong et al., 2016). The

finding of this study are in line with those of previous studies. For instance, a study

by Beljouw et al. (2014) concluded that depression was a consequence of loneliness

for community-dwelling elderly people and severe cases of depression were found

77 | P a g e

among older adults who felt lonely. Further evidence comes from a study by Barg et

al. (2006) regarding perception of loneliness among both depressed and not

depressed older adults which reported that most of the older adults perceived

loneliness as a precursor to a subsequent depression. Similar results were reported by

Singh and Mishra (2009) that loneliness was significantly associated with depression.

Gan, Xie, Duan, Deng, and Yu (2015) in their six-month longitudinal study for

Chinese older adults also showed that loneliness at baseline was a predictor of

depression at the follow up wave.

c)Insomnia

Insomnia was found to be another predictor of depression in this study, it was highly

significant throughout all 6 waves of ELSA (p<0.001). The finding that insomnia

was significantly associated with depression was not surprising as literature have

demonstrated that insomnia is one of the most common feature observed in

depressed people, thus, often it is looked as a symptom of depression (Benca &

Peterson, 2008). Moreover, studies have also highlighted that how insomnia is of

central importance for the onset of new depressive episodes (Mcnamara, 2006). The

finding of this study is in agreement with the results obtained by Livingston, Blizarda

and Mann (1993) where the authors found that insomnia was the best predictor of

depression for older adults living in London. They also reported that in the presence

of insomnia in their multivariate model, the traditional significant risk factors

including demographic factors and long standing illness became insignificant.

Similar results were replicated by (Nolan, 2009) who confirmed that insomnia was

an independent predictor not only for the development of depression but also for

recurrent depression among elderly people. Likewise, a research conducted by

Riemann and Voderholzer (2003) identified that insomniac patients became

depressed in the follow-up interviews, which showed insomnia as a predictor of

depression.

d)Financial Strain

Financial strain came out to be another predictor of depression in this study as it was

significantly associated with depression throughout all 6 waves of ELSA. A

significant association of economic hardship with depression in older adults was also

78 | P a g e

found by Pudrovska, Schieman, Pearlin, and Nguyen (2005). This view is supported

by Sharma, Satija, and Nathawat (1985) in their study about various life events that

relates to depression in older adults, in which they concluded that life event related to

financial hardship was the most associated event with depression. Similar results

were replicated by Aranda and Lincoln (2011) where they demonstrated that

financial strain was an independent predictor of depression. In the same vein, various

other studies have provided convincing evidence to suggest financial strain to be

significantly associated with depression (Krause, 1987; Zimmerman & Katon, 2005;

Price, Choi, & Vinokur, 2002). Researchers have emphasized that among different

socio-economic factors that relate to depression, financial strain is the strongest

among all of them and financial strain can even mediate the significant association of

other socio-economic factors with depression (Kessler, Turner, & House, 1987;

Whelan, 1993). A longitudinal study to measure impact of economic crisis on older

adults by Sargent-Cox, Butterworth, & Anstey (2011) also suggested that older

adults who indicated an impact due to economic crisis were more associated with

depressive symptoms.

5.2. Insignificant factors - Not at all significant

a). Long standing illness/disability/infirmity

Contrary to expectations, this study did not find a significant association between

long standing illness and depression. Much of the literature has focussed on existence

of a strong significant association between long standing illness/disability/infirmity

with depression (Dickens et al., 2011; Rifel, Švab, Pavlič, King, & Nazareth, 2010)

but not all, for instance, H. Lee, Hahn, Shim, Kwon, & Jeong, 2013 showed that long

standing illness was not associated with depression in older adults, and the number of

physical illness present was also not associated with depression, similar to the

findings of this study. Similarly, of particular relevance to the findings of present

study is the work of Alpass & Neville (2003) where they concluded that physical

disability and presence of long standing illness was not associated with depression.

Using similar methods, where they considered self-rated health, physical illness and

long term health problems in a multivariate model, the authors discovered that

depression was associated with only self-rated health as individuals who reported

poor self-rated health had experience more depression compared to other individuals.

79 | P a g e

Similarly, a study by Beekman et al. (1997) concluded that long standing illness was

associated only with minor depression and not with major depression among elderly.

b). Marital status

In this study, marital status was not a risk factor for depression, as it remained

insignificant in subset data and at each wave of ELSA. These findings match those

observed in earlier studies. Jang et al., (2009) reported that in older adults the

proportion of depressed people for both genders were not associated with marital

status. Similar results were replicated by Markides & Farrell (1985) who reported

that marital status lost its significance when other factors were taken into account. A

possible explanation for this finding is that a great deal of research that has focussed

on relationship between marital status and depression have reported different results

about association of marital status with depression, few have reported married were

less depressed (Etaugh & Malstrom, 1981) on the other hand, it has been published

that negative events that occur in marital life leads to more depression among women

compared to any other marital status (Aseltine & Kessler, 1993). Few studies have

reported widowed to be the most depressed group among elderly people (Harlow,

1991) whereas Etaugh & Malstrom, 1981 offered valid counterarguments that

widowed were less depressed than women with marital status as divorce. Hence, it

seems possible that older adult’s marital status doesn’t have a significant relation

with depression

5.3. Mostly insignificant factors

a) Gender

In all 6 waves of ELSA, the proportion of depressed females was consistently higher

than males. However, the results of logistic regression in this study showed that

gender was not a significant risk factor for depression for most of the waves when it

was added in a multivariate model along with demographic, health, and other factors

related to depression. These results are consistent with those of Tazelaar et al. (2008)

who demonstrated that the effect of gender on depression was not significant after

addition of variables related to health and loneliness in their study. Similar findings

were replicated by Dessoki, Moussa and Nasr (2012) who confirmed the

80 | P a g e

insignificant association between gender and depression in their study, although, they

found that depression in females was more associated with the feeling of

worthlessness and suffering. In this study, however, gender was significant at wave 3

and wave 4 which may be explained by the fact that females have a higher tendency

to report symptoms of anxiety, stress and depression (Weissman, 1977). Another

possible explanation for this if considered from a biological point of view is, it is

more mood changes in women due to reproductive cycles that make them more

vulnerable to depression (Hoeksema, 1987) and when considered from a

psychosocial point of view, the higher risk of depression in females is due to both

low social status attached with female gender and a greater concern for relationships

(Weissman, 1977). There are, yet, several other possible explanations for this result.

b) Age

The results obtained for age were similar as obtained for gender in sense that it was

significant at only one of the wave and remained insignificant for all other waves.

This finding of age not being significant for most of the waves is consistent with that

of Blazer, Burchett, Service, and George (1991) who reported that increased age was

significantly associated with depression but lost its significance when other factors

related to health, social support and financial were considered. Similarly, study by

Beekman (1995) also showed that both gender and age were not risk factors for

depression in elderly. A possible explanation for this inconsistency of age as risk

factor supports the idea of Jorm (2000) who concluded that there is a disagreement

between whether depression decreases in older adults due to decreased need of

emotional support and increased control over emotions, or whether it increases due to

greater vulnerability of getting alone, and decreased health status. However, its

significance at two of the waves support the ideas of Yang (2007) that with age,

depression symptoms are obvious, due to presence of different factors in old age

which are independently associated with depression. Similarly, Stordal, Mykletun,

and Dahl (2003) also found age to be significantly associated with depression.

c). Alcohol

Alcohol consumption was not a risk factor for depression as it remained insignificant

in subset data as well as in all 6 waves of ELSA other than in wave 6. This finding is

81 | P a g e

consistent with results obtained by van Gool (2003) that alcohol consumption was

not significantly associated with older adults. Similarly, Weyerer et al., (2013) also

emphasised that depression was not associated with alcohol consumption for elderly

people. In the same vein, Barry, Fleming, Manwell, Copeland, and Appel (1998) also

reported that neither the volume nor the frequency of alcohol consumption was

associated with depression among elderly. Alcohol consumption as insignificant for

each of the wave except at wave 6 was not a very expected finding because a great

deal of literature has also highlighted a strong association between alcohol

consumption and depression (Bekaroğlu, Uluutku, Tanriöver, & Kirpinar, 1991;

Rodgers et al., 2000; Perreira & Sloan, 2002; Aihara, Minai, Aoyama, &

Shimanouchi, 2010). A possible explanation for this finding is may be the lack of a

strong evidence from the literature about whether this association exists or not, a

number of studies differ in their interpretation of relationship between alcohol

consumption and depression in older adults. It is difficult to guess about whether this

association is based on the data that authors use for their research or whether it

actually exists.

5.4. Not associated factor

a) Children

A very strange finding was that the factor - children was not significant even at

bivariate analysis (from the results of chi-square test). This study, hence, suggest that

in the UK, depression in older adults is not significantly associated with the “children”

factor. Our finding evidently support the results published from the research

performed by Vikström et al. (2011) that no difference was found in depression

among older adults based upon whether they have children or not. The relationship

between childlessness in later life with depression was also analysed by Cox (1998)

where he demonstrates that childlessness was not associated with depression among

the elderly even after controlling all other factors. A possible explanation for children

factor not having associated with depression might be that older adults having

children and not getting the required support is also associated with higher levels of

depression (Djundeva, Mills, Wittek, & Steverink, 2015). This idea is further

supported by research performed by COX (2002) where he identified that the highest

levels of depression were present among older adults that shared a poor quality

relationship with their children.

82 | P a g e

Chapter Six: Conclusion and Recommendations

6.1. Conclusion

The identified predictors from this study were self-rated health, loneliness, insomnia

and financial strain. Insignificance of demographic factors (age and gender for most

of the waves and marital status for all 6 waves and no association altogether at bi-

variate analysis for the factor children) showed that depression cannot be attributed

to only traditional significant risk factors that have been discussed in the published

literature. The finding, self-rated health to be significantly associated with depression

whereas insignificance of long term illness with depression need further investigation

whether depression is associated with how the respondents think their health is or is

it based on actual presence of some illness/disability/infirmity. Also, alcohol

consumption in older adults was found not be associated with depression. Although,

this study identified predictors of depression within the ELSA, it cannot ensure these

are the factors that cause depression in older adults, it can only be considered that

these factors are strongly associated with depression.

6.2. Limitations of this study

a). From the data perspective

Depression variable was measured based on the response to the question “Much of

the time during the past week, you felt depressed?”. This way of measurement has its

own limitations as follows. First, not all older adults can remember and recall how

they felt last week, and thus it includes a possibility of memory bias. Second, it is not

necessary that a respondent who felt depressed last week actually had a clinically

recognised depression. Third, a person may have experienced sad mood due to a

stressful event or other difficult circumstances which may seems to them as

depression. Third, as discussed in the literature review that many older adults are not

able to detect depression due to different other health problems, stigma attached with

depression, and few considering it as a part of normal ageing. The use of CES-D

scale (Center for Epidemiologic Studies Depression Scale) to measure depression

would have been more useful to obtain accurate results.

83 | P a g e

Another limitation of this study is that a single question was asked to measure long

term illness, disability and infirmity. Whereas in current and past published literature

each of the factor has been considered independently by the researchers where they

have found different associations of these three factors with depression. So the

finding of this study doesn’t actually clear whether it is disability, long term illness

or infirmity that was not at all significant.

Another limitation of this study is that it has analysed data of a longitudinal survey,

and longitudinal surveys have their own drawbacks as comparison is made on the

same basis for each wave even though respondents at each wave are dropped out or

new ones are added, the problem of attrition that is accompanied with every

longitudinal study.

b). From the methods perspective

In logistic regression, three models were used to look for which factors predicts

depression. Although, the models were divided into demographic, health and other

factors. It has its own limitation as it has not considered adding self-rated health and

presence of long standing health problem in different models to check whether long

standing illness was a predictor for depression when self-rated health was not taken

into account and what correlation exists between self-rated health and long standing

illness. Similarly, for other variables.

In this study only two of the data mining techniques have been used to identify the

predictive nature of the factors. But, may be different algorithms such as Baseline

classifier (ZeroR), Artificial neural networks, Naive Bayes may have produced

different results, better or worse than these obtained results.

c)From the results perspective

The results obtained for predictors of depression should be interpreted as predictors

of depression amongst the ten factors that were considered in this present study. As

there were various other factors that were identified to be risk factors of depression

from the published literature but were not included in this study considering the

scope of the present study.

84 | P a g e

6.3. Strengths of this study

This study has tried to find predictors of depression by considering each of the wave

to ensure that it reports the right results about predictors of depression in terms of the

most significant ones i.e. the factors that came out to be the predictors for each wave

of ELSA.

Additionally, data filtered for people who were either depressed or not depressed

throughout all waves of ELSA provides further evidence about what predicts

depression in older adults of England based on these ten factors.

A good accuracy rate produced by data mining techniques have further shown the

predictive abilities of these factors to predict depression.

Status of each factor based on mode value for the respondents who were depressed

throughout all waves of ELSA further provide better understanding about these ten

factors and in what state were they present for the respondents who were depressed

through these 11 years of ELSA survey.

6.4. Recommendations for Future Research

Various areas that could further be explored while finding the predictors of

depression within the ELSA that were currently outside the scope of this study are as

follows. First, in this study the variables that were added were present throughout

each of the wave but the ELSA contains some additional variables and sub modules

at different waves such as “effort and reward module” at wave 2; “risk module”

related questions that were asked only at wave 5; life history questions at wave 3;

variables related to sexual activity present only at wave 6. It would be interesting to

assess the effects of these variables from different waves along with traditional risk

factors such as demographic factors to identify which of the variables relate more to

depression in older adults.

It is recommended that further research be undertaken using only data mining

techniques to predict depression in older adults. As much more variables could be

added in the data mining software to first select which one of them are the most

relevant using the select attributes functionality and by selecting appropriate attribute

evaluator and a search method; for ELSA the best would be :cfsSubsetEval”

85 | P a g e

(attribute evaluator) and “Best First”(method) and then these factors could be further

used to look at the accuracy rates being produced by different data mining algorithms

(mainly SVM, artificial neural networks, Random Forests, J48, Random tree, Naive

bayes, and ZeroR).

The way tree diagram was produced in this study to look for how people changed

their status of depression from one wave to another, it would be very interesting if a

longitudinal analysis is carried out to assess what factors were changed from one

wave to another when a change in depression status was observed.

References

All, M. A. H., Uysse, D. A. J. B., Owell, P. E. D. N., Ofzinger, E. R. I. C. A. N.,

Ouck, P. A. H., Eynolds, C. H. F. R., & Upfer, D. A. J. K. (2000). Symptoms of

Stress and Depression as Correlates of Sleep in Primary Insomnia, 230, 227–

230.

Alpass, F. M., & Neville, S. (2016). Loneliness , health and depression in older

males, 7863(April). http://doi.org/10.1080/1360786031000101193

Anderson, D. N. (2001). Treating depression in old age : the reasons to be positive,

13–17.

Aihara, Y., Minai, J., Aoyama, A., & Shimanouchi, S. (2010). Depressive symptoms

and past lifestyle among Japanese elderly people. Community Mental Health

Journal, 47(2), 186–193. doi:10.1007/s10597-010-9317-1

Akhtar-Danesh, N., & Landeen, J. (2007). Relation between depression and

sociodemographic factors. International Journal of Mental Health Systems, 1(1),

4. doi:10.1186/1752-4458-1-4

Alexopoulos, G. S. (2005). Depression in the elderly. The Lancet, 365(9475), 1961–

1970. doi:10.1016/s0140-6736(05)66665-2

Alpass, F. M., & Neville, S. (2003). Loneliness, health and depression in older males.

Aging & Mental Health, 7(3), 212–216. doi:10.1080/1360786031000101193

86 | P a g e

Ambresin, G., Chondros, P., Dowrick, C., Herrman, H., & Gunn, J. M. (2014). Self-

rated health and long-term prognosis of depression. The Annals of Family

Medicine, 12(1), 57–65. doi:10.1370/afm.1562

Aneshensel, C. S., Frerichs, R. R., & Huba, G. J. (1984). Depression and physical

illness: A Multiwave, Nonrecursive causal model. Journal of Health and Social

Behavior, 25(4), 350. doi:10.2307/2136376

Aranda, M. P., & Lincoln, K. D. (2011). Financial strain, negative interaction, coping

styles, and mental health among low-income Latinos. Race and Social Problems,

3(4), 280–297. doi:10.1007/s12552-011-9060-4

Aseltine, R. H., & Kessler, R. C. (1993). Marital disruption and depression in a

community sample. Journal of Health and Social Behavior, 34(3), 237.

doi:10.2307/2137205

Aylaz, R., Aktürk, Ü., Erci, B., Öztürk, H., & Aslan, H. (2012). Relationship between

depression and loneliness in elderly and examination of influential factors.

Archives of Gerontology and Geriatrics, 55(3), 548–554. doi:

10.1016/j.archger.2012.03.006

Banerjee, S. (2014). Multimorbidity — older adults need health care that can count,

6736(14), 587–589. http://doi.org/10.1016/S0140-6736(14)61596-8

Blazer, D., Burchett, B., Service, C., & George, L. K. (1991). The Association of

Age and Depression Among the Elderly : An Epidemiologic Exploration, 46(6).

Blazer, D. G. (2003). Depression in Late Life : Review and Commentary, 58(3),

249–265.

Blazer, D., Hughes, D. C., & George, L. K. (1987). The Epidemiology of Depression

in an Elderly Community Population 1, 27(3), 281–287.

Badawi, G., Pagé, V., Smith, K. J., Gariépy, G., Malla, A., Wang, J., … Schmitz, N.

(2013). Self-rated health: A predictor for the three-year incidence of major

depression in individuals with type II diabetes. Journal of Affective Disorders, 145(1),

100–105. doi: 10.1016/j.jad.2012.07.018

Barg, F. K., Huss-Ashmore, R., Wittink, M. N., Murray, G. F., Bogner, H. R., & Gallo, J.

J. (2006). A mixed-methods approach to understanding loneliness and

87 | P a g e

depression in older adults. The Journals of Gerontology Series B: Psychological

Sciences and Social Sciences, 61(6), S329–S339. doi:10.1093/geronb/61.6. s329

Beekman, A. T. F., Deeg, D. J. H., van Tilburg, T., Smit, J. H., Hooijer, C., & van Tilburg,

W. (1995). Major and minor depression in later life: A study of prevalence and

risk factors. Journal of Affective Disorders, 36(1-2), 65–75. doi:10.1016/0165-

0327(95)00061-5

Bekaroğlu, M., Uluutku, N., Tanriöver, S., & Kirpinar, I. (1991). Depression in an

elderly population in turkey. Acta Psychiatrica Scandinavica, 84(2), 174–178.

doi:10.1111/j.1600-0447. 1991.tb03124.x

Bekhet, A. K., & Zauszniewski, J. A. (2012). Mental health of elders in retirement

communities: Is loneliness a key factor? Archives of Psychiatric Nursing, 26(3),

214–224. doi: 10.1016/j.apnu.2011.09.007

Benca, R. M., & Peterson, M. J. (2008). Insomnia and depression. Sleep Medicine, 9,

S3–S9. doi:10.1016/s1389-9457(08)70010-8

Benton, T., Staab, J., & Evans, D. L. (2007). Medical Co-Morbidity in Depressive

disorders. Annals of Clinical Psychiatry, 19(4), 289–303.

doi:10.1080/10401230701653542

Berger, B. D., & Adesso, V. J. (1991). Gender differences in using alcohol to cope

with depression. Addictive Behaviors, 16(5), 315–327. doi:10.1016/0306-

4603(91)90024-c

Blazer, D., Burchett, B., Service, C., & George, L. K. (1991). The association of age

and depression among the elderly: An epidemiologic exploration. Journal of

Gerontology, 46(6), M210–M215. doi:10.1093/geronj/46.6.m210

Blazer, D., Hughes, D. C., & George, L. K. (1987). The Epidemiology of depression in

an elderly community population. The Gerontologist, 27(3), 281–287.

doi:10.1093/geront/27.3.281

Borawski, E. A., Kinney, J. M., & Kahana, E. (1996). The meaning of older adults’

health appraisals: Congruence with health status and determinant of mortality.

The Journals of Gerontology Series B: Psychological Sciences and Social Sciences,

51B(3), S157–S170. doi:10.1093/geronb/51b.3.s157

Bourassa, K. J., Memel, M., Woolverton, C., & Sbarra, D. A. (2015). Social

participation predicts cognitive functioning in aging adults over time:

88 | P a g e

Comparisons with physical health, depression, and physical activity. Aging &

Mental Health. doi:10.1080/13607863.2015.1081152

Bower, B. (1986). Treating depression: Can we talk? Science News, 129(21), 324.

doi:10.2307/3970683

Broudeur, H., Hurrell, D., Stepinska, M., Fluffy, P. A., & Houxou. (2014, October).

ELSA - English longitudinal study of Ageing. Retrieved August 25, 2016, from

http://www.elsa-project.ac.uk/

Bruce, M. L., & Kim, K. M. (1992). Differences in the effects of divorce on major

depression in men and women. American Journal of Psychiatry, 149(7), 914–917.

doi:10.1176/ajp.149.7.914

Bulloch, A., Lavorato, D., Williams, J., & Patten, S. (2012). ALCOHOL CONSUMPTION

AND MAJOR DEPRESSION IN THE GENERAL POPULATION: THE CRITICAL

IMPORTANCE OF DEPENDENCE. Depression and Anxiety, 29(12), 1058–1064.

doi:10.1002/da.22001

Butterworth, P., Rodgers, B., & Windsor, T. D. (2009). Financial hardship, socio-

economic position and depression: Results from the PATH through life survey.

Social Science & Medicine, 69(2), 229–237. doi:

10.1016/j.socscimed.2009.05.008

Byrne, E. J. (1994). Diagnosis and treatment of depression in late life. Edited by Lon

S. Schneider, Charles F. Reynolds 111, Barry D. Lebowitz and Arnold Friedhoff.

American psychiatric press, 1994. No. Of pages: 416. Price: £41.95. International

Journal of Geriatric Psychiatry, 9(7), 592–592. doi:10.1002/gps.930090716

Cairney, J., & Wade, T. J. (2002). The influence of age on gender differences in

depression. Social Psychiatry and Psychiatric Epidemiology, 37(9), 401–408.

doi:10.1007/s00127-002-0569-0

Carruthers, H. R., Morris, J., Tarrier, N., & Whorwell, P. J. (2010). The Manchester

color wheel: Development of a novel way of identifying color choice and its

validation in healthy, anxious and depressed individuals. BMC Medical Research

Methodology, 10(1), . doi:10.1186/1471-2288-10-12

Chang, K. J., Son, S. J., Lee, Y., Back, J. H., Lee, K. S., Lee, S. J., … Hong, C. H. (2014).

Perceived sleep quality is associated with depression in a Korean elderly

89 | P a g e

population. Archives of Gerontology and Geriatrics, 59(2), 468–473. doi:

10.1016/j.archger.2014.04.007

Chou, K.-L., & Chi, I. (2004). Prevalence and correlates of depression in Chinese

oldest-old. International Journal of Geriatric Psychiatry, 20(1), 41–50.

doi:10.1002/gps.1246

Chow, S. K. Y., & Chan, W. C. (2010). Research article: Depression: Problem-solving

appraisal and self-rated health among Hong Kong Chinese migrant women.

Nursing & Health Sciences, 12(3), 352–359. doi:10.1111/j.1442-2018.2010.

00537.x

Christensen, K., Doblhammer, G., Rau, R., & Vaupel, J. W. (2009). Ageing

populations: The challenges ahead. The Lancet, 374(9696), 1196–1208.

doi:10.1016/s0140-6736(09)61460-4

Comstock, G. W., & Helsing, K. J. (1977). Symptoms of depression in two

communities. Psychological Medicine, 6(04), 551.

doi:10.1017/s0033291700018171

Connidis, I. A., & McMullin, J. A. (1993). To have or have not: Parent status and the

subjective well-being of older men and women. The Gerontologist, 33(5), 630–

636. doi:10.1093/geront/33.5.630

Conwell, Y., Duberstein, P. R., & Caine, E. D. (2002). Risk factors for suicide in later

life. Biological Psychiatry, 52(3), 193–204. doi:10.1016/s0006-3223(02)01347-1

Craft, B. J., Johnson, D. R., & Ortega, S. T. (1998). Rural-urban women’s experience

of symptoms of depression related to economic hardship. Journal of Women &

Aging, 10(3), 3–18. doi:10.1300/j074v10n03_02

Crome, I., & Crome, P. (2007). Moderate alcohol consumption in older adults is

associated with better cognition and well-being than abstinence. Age and

Ageing, 37(1), 120–121. doi:10.1093/ageing/afm150

Culbertson, F. M. (1997). Depression and gender: An international review. American

Psychologist, 52(1), 25–31. doi:10.1037/0003-066x.52.1.25

Campayo, A., & Gómez-biel, C. H. (2011). Diabetes and Depression, 26–30.

http://doi.org/10.1007/s11920-010-0165-z

Coussement, K., Bossche, F. A. M. Van Den, & Bock, K. W. De. (2014). Data

90 | P a g e

accuracy ’ s impact on segmentation performance : Benchmarking RFM

analysis , logistic regression , and decision trees. Journal of Business Research,

67(1), 2751–2758. http://doi.org/10.1016/j.jbusres.2012.09.024

Damián, J., Pastor-Barriuso, R., & Valderrama-Gama, E. (2008). Factors associated

with self-rated health in older people living in institutions. BMC Geriatrics, 8(1),

5. doi:10.1186/1471-2318-8-5

Dessoki, H., Moussa, F., & Nasr, M. (2011). P02-236 - gender differences in elderly

patientswith depression. European Psychiatry, 26, 832. doi:10.1016/s0924-

9338(11)72537-8

Dickens, C., Coventry, P., Khara, A., Bower, P., Mansell, W., & Bakerly, N. D. (2011).

Perseverative negative cognitive processes are associated with depression in

people with long-term conditions. Chronic Illness, 8(2), 102–111.

doi:10.1177/1742395311433058

Drageset, J., Espehaug, B., & Kirkevold, M. (2012). The impact of depression and

sense of coherence on emotional and social loneliness among nursing home

residents without cognitive impairment - a questionnaire survey. Journal of

Clinical Nursing, 21(7-8), 965–974. doi:10.1111/j.1365-2702.2011. 03932.x

Druss, B. G., Hwang, I., Petukhova, M., Sampson, N. A., Wang, P. S., & Kessler, R. C.

(2008). Impairment in role functioning in mental and chronic medical disorders

in the United States: Results from the national Comorbidity survey replication.

Molecular Psychiatry, 14(7), 728–737. doi:10.1038/mp.2008.13

Dendukuri, N., & Cole, M. G. (2001). Risk Factors for Depression Among Elderly

Community subjects : A Systematic Review and Meta-Analysis, (12), 1147–

1156.

Djundeva, M., Mills, M., Wittek, R., & Steverink, N. (2015). Receiving Instrumental

Support in Late Parent–Child Relationships and Parental Depression. The

Journals of Gerontology Series B: Psychological Sciences and Social Sciences,

70(6), 981–994. http://doi.org/10.1093/geronb/gbu136

Etaugh, C., & Malstrom, J. (1981). The effect of marital status on person perception.

Journal of Marriage and the Family, 43(4), 801. doi:10.2307/351337

91 | P a g e

Evenson, R. J., Simon, R. W., Beth, S., Glass, J., Thoits, P., & Hughes, M. (2005).

Clarifying the Relationship Between Parenthood and Depression *

FARAKHAN, A., LUBIN, B., & O’CONNOR, W. A. (1984). LIFE SATISFACTION AND

DEPRESSION AMONG RETIRED BLACK PERSONS. Psychological Reports, 55(2),

452–454. doi:10.2466/pr0.1984.55.2.452

Faravelli, C., Alessandra Scarpato, M., Castellini, G., & Lo Sauro, C. (2013). Gender

differences in depression and anxiety: The role of age. Psychiatry Research,

210(3), 1301–1303. doi: 10.1016/j.psychres.2013.09.027

Fishleder, S., Schonfeld, L., Corvin, J., Tyler, S., & VandeWeerd, C. (2015). Drinking

behavior among older adults in a planned retirement community: Results from

the villages survey. International Journal of Geriatric Psychiatry, 31(5), 536–543.

doi:10.1002/gps.4359

Ford, D. E. (1989). Epidemiologic study of sleep disturbances and psychiatric

disorders. JAMA, 262(11), 1479. doi:10.1001/jama.1989.03430110069030

Foster, C. J., Barkus, E., Yavorsky, C., Foster, E. J., Barkus, E., & Yavorsky, C.

(2016). Understanding and Using Advanced Statistics Logistic Regression, 57–

70.

Franke, T. M., Ho, T., & Christie, C. A. (2012). The Chi-Square Test : Often Used

and More Often Misinterpreted. http://doi.org/10.1177/1098214011426594

Gambhir, I. S., Chakrabarti, S. S., Sharma, A. R., & Saran, D. P. (2014). Insomnia in

the elderly—A hospital-based study from north India. Journal of Clinical

Gerontology and Geriatrics, 5(4), 117–121. doi: 10.1016/j.jcgg.2014.05.005

Gan, P., Xie, Y., Duan, W., Deng, Q., & Yu, X. (2015). Rumination and loneliness

independently predict Six-Month later depression symptoms among Chinese

elderly in nursing homes. PLOS ONE, 10(9), e0137176. doi:

10.1371/journal.pone.0137176

Glenn, N. D. (1975). The contribution of marriage to the psychological well-being of

males and females. Journal of Marriage and the Family, 37(3), 594.

doi:10.2307/350523

92 | P a g e

Graham, K., & Schmidt, G. (1999). Alcohol use and psychosocial well-being among

older adults. Journal of Studies on Alcohol, 60(3), 345–351.

doi:10.15288/jsa.1999.60.345

Gureje, O., Ademola, A., & Olley, B. O. (2008). Depression and disability:

Comparisons with common physical conditions in the Ibadan study of aging.

Journal of the American Geriatrics Society, 56(11), 2033–2038.

doi:10.1111/j.1532-5415.2008. 01956.x

Goldstein, B., Rosselli, F., Goldstein, B., & Rosselli, F. (2016). Etiological

paradigms of depression : The relationship between perceived causes ,

empowerment , treatment preferences , and stigma Etiological paradigms of

depression : The relationship between perceived causes , empowerment ,

treatment, 8237(April). http://doi.org/10.1080/09638230310001627919

Han, B. (2002). Depressive symptoms and self-rated health in community-dwelling

older adults: A longitudinal study. Journal of the American Geriatrics Society,

50(9), 1549–1556. doi:10.1046/j.1532-5415.2002. 50411.x

Harlow, S. D. (1991). A longitudinal study of the prevalence of Depressive

Symptomatology in elderly widowed and married women. Archives of General

Psychiatry, 48(12), 1065. doi:10.1001/archpsyc.1991.01810360029004

Haseen, F., & Prasartkul, P. (2011). Predictors of depression among older people

living in rural areas of Thailand. Bangladesh Medical Research Council Bulletin,

37(2), doi:10.3329/bmrcb. v37i2.8434

Hawkes, N. (2012). Susceptibility of people with long term illness to depression and

anxiety is not recognised, report says. BMJ, 344(feb08 2), e950–e950.

doi:10.1136/bmj. e950

Holtfreter, K., Reisig, M. D., & Turanovic, J. J. (2015). Depression and infrequent

participation in social activities among older adults: The moderating role of high-

quality familial ties. Aging & Mental Health.

doi:10.1080/13607863.2015.1099036

Holvast, F., Burger, H., de Waal, M. M. W., van Marwijk, H. W. J., Comijs, H. C., &

Verhaak, P. F. M. (2015). Loneliness is associated with poor prognosis in late-life

depression: Longitudinal analysis of the Netherlands study of depression in

93 | P a g e

older persons. Journal of Affective Disorders, 185, 1–7. doi:

10.1016/j.jad.2015.06.036

Hosseini, H. (2014). Aging and the rising costs of healthcare in the United States:

Can there be a solution? Ageing International, 40(3), 229–247.

doi:10.1007/s12126-014-9209-8

Hamer, M., Batty, G. D., & Kivimaki, M. (2012). Risk of future depression in people

who are obese but metabolically healthy : the English longitudinal study of

ageing. Molecular Psychiatry, 17(9), 940–945.

http://doi.org/10.1038/mp.2012.30

Hill, C. V, Pérez-stable, E. J., Anderson, N. A., & Bernard, M. A. (2015). T he N

ational I nstitute on A ging H ealth D isparities R esearch F ramework, 25(3),

245–254.

Hodges, S. (2002). Mental Health , Depression , and Dimensions of Spirituality and

Religion, 9(2).

Jacobson, A. M. (1993). Depression and diabetes. Diabetes Care, 16(12), 1621–1623.

doi:10.2337/diacare.16.12.1621

Jang, S.-N., Kawachi, I., Chang, J., Boo, K., Shin, H.-G., Lee, H., & Cho, S. (2009).

Marital status, gender, and depression: Analysis of the baseline survey of the

Korean longitudinal study of Ageing (KLoSA). Social Science & Medicine, 69(11),

1608–1615. doi: 10.1016/j.socscimed.2009.09.007

Jang, Y., Park, N. S., Kang, S.-Y., & Chiriboga, D. A. (2013). Racial/ethnic differences

in the association between symptoms of depression and self-rated mental

health among older adults. Community Mental Health Journal, 50(3), 325–330.

doi:10.1007/s10597-013-9642-2

Jaremka, L. M., Andridge, R. R., Fagundes, C. P., Alfano, C. M., Povoski, S. P., Lipari, A.

M., … Kiecolt-Glaser, J. K. (2014). Pain, depression, and fatigue: Loneliness as a

longitudinal risk factor. Health Psychology, 33(9), 948–957.

doi:10.1037/a0034012

Johnson, I. (2000). Alcohol problems in old age: A review of recent epidemiological

research. International Journal of Geriatric Psychiatry, 15(7), 575–581.

doi:10.1002/1099-1166(200007)15:7<575: aid-gps151>3.0.co;2-0

94 | P a g e

Joiner, T. E., & Blalock, J. A. (1995). Gender differences in depression: The role of

anxiety and generalized negative affect. Sex Roles, 33(1-2), 91–108.

doi:10.1007/bf01547937

Jokela, M., Batty, G. D., & Kivimäki, M. (2013). Ageing and the prevalence and

treatment of mental health problems. Psychological Medicine, 43(10), 2037–

2045. doi:10.1017/s0033291712003042

JORM, A. F. (2000). Does old age reduce the risk of anxiety and depression? A

review of epidemiological studies across the adult life span. Psychological

Medicine, 30(1), 11–22. doi:10.1017/s0033291799001452

Jorm, A. F. (2000). Does old age reduce the risk of anxiety and depression ? A

review of epidemiological studies across the adult life span, 11–22.

Kader Maideen, S. F., Mohd Sidik, S., Rampal, L., & Mukhtar, F. (2015). Prevalence,

associated factors and predictors of anxiety: A community survey in Selangor,

Malaysia. BMC Psychiatry, 15(1), . doi:10.1186/s12888-015-0648-x

Kamiya, Y., Doyle, M., Henretta, J. C., & Timonen, V. (2013). Depressive symptoms

among older adults: The impact of early and later life circumstances and marital

status. Aging & Mental Health, 17(3), 349–357.

doi:10.1080/13607863.2012.747078

Karpansalo, M. (2005). Depression and early retirement: Prospective population

based study in middle aged men. Journal of Epidemiology & Community Health,

59(1), 70–74. doi:10.1136/jech.2003.010702

Katsumata, Y., Arai, A., Ishida, K., Tomimori, M., Denda, K., & Tamashiro, H. (2005).

Gender differences in the contributions of risk factors to depressive symptoms

among the elderly persons dwelling in a community, Japan. International Journal

of Geriatric Psychiatry, 20(11), 1084–1089. doi:10.1002/gps.1403

Kessler, R. C., Turner, J. B., & House, J. S. (1987). Intervening processes in the

relationship between unemployment and health. Psychological Medicine, 17(04),

949. doi:10.1017/s0033291700000763

Kim, J.-M., Shin, I.-S., Yoon, J.-S., & Stewart, R. (2002). Prevalence and correlates of

late-life depression compared between urban and rural populations in Korea.

95 | P a g e

International Journal of Geriatric Psychiatry, 17(5), 409–415.

doi:10.1002/gps.622

Kivelá, S.-L., & Pahkala, K. (2001). Depressive disorder as a predictor of physical

disability in old age. Journal of the American Geriatrics Society, 49(3), 290–296.

doi:10.1046/j.1532-5415.2001. 4930290.x

Koenig, H. G. (1988). Depression in elderly hospitalized patients with medical illness.

Archives of Internal Medicine, 148(9), 1929–1936.

doi:10.1001/archinte.148.9.1929

Koropeckyj-Cox, T. (1998). Loneliness and depression in middle and old age: Are the

childless more vulnerable? The Journals of Gerontology Series B: Psychological

Sciences and Social Sciences, 53B (6), S303–S312.

doi:10.1093/geronb/53b.6.s303

Koropeckyj-Cox, T. (2002). Beyond parental status: Psychological well-being in

middle and old age. Journal of Marriage and Family, 64(4), 957–971.

doi:10.1111/j.1741-3737.2002. 00957.x

Krause, N. (1987). Chronic financial strain, social support, and depressive symptoms

among older adults. Psychology and Aging, 2(2), 185–192. doi:10.1037/0882-

7974.2.2.185

Kronmüller, K.-T., Backenstrass, M., Victor, D., Postelnicu, I., Schenkenbach, C., Joest,

K., … Mundt, C. (2011). Quality of marital relationship and depression: Results of

a 10-year prospective follow-up study. Journal of Affective Disorders, 128(1-2),

64–71. doi: 10.1016/j.jad.2010.06.026

Lebowitz, B. D. (1997). Diagnosis and treatment of depression in late life. JAMA,

278(14), 1186. doi:10.1001/jama.1997.03550140078045

Kadane, J. B. (2016). Review : Descriptive Statistics Authors ( s ): Joseph B .

Kadane Review by : Joseph B . Kadane Source : Science , New Series , Vol .

200 , No . 4338 ( Apr . 14 , 1978 ), p . 195 Published by : American Association

for the Advancement of Science Stable URL, 200(4338).

Kim H. K., & McKenry, P. C. (2002). The Relationship Between Marriage and

Psychological Well-being: A Longitudinal Analysis. Journal of Family Issues,

23(8), 885–911. http://doi.org/10.1177/019251302237296

96 | P a g e

Kingsford, C., & Salzberg, S. L. (2008). What are decision trees ? a, 26(9), 1011–

1013.

Lee, H. (2014). The impact of social activity on life satisfaction and depression of

community-dwelling elderly:comparing living arrangement. Journal of

community welfare, 48(1), doi:10.15300/jcw.2014.48.1.269

Lee, H., Hahn, S., Shim, S., Kwon, Y., & Jeong, H. (2013). P.2.b.057 prevalence of

depression in elderly patients with chronic physical illness. European

Neuropsychopharmacology, 23, S353. doi:10.1016/s0924-977x (13)70556-x

Lochen, M.., & Rasmussen, K. (1996). Palpitations and lifestyle: Impact of

depression and self-rated health. The Nordland health study. Scandinavian

Journal of Public Health, 24(2), 140–144. doi:10.1177/140349489602400209

Loder, B. (2009). Depression in later life: A personal account. Quality in Ageing and

Older Adults, 10(1), 47–48. doi:10.1108/14717794200900008

Lopez, A. D., & Murray, C. C. J. L. (1998). The global burden of disease, 1990–2020.

Nature Medicine, 4(11), 1241–1243. doi:10.1038/3218

Lue, B.-H., Chen, L.-J., & Wu, S.-C. (2010). Health, financial stresses, and life

satisfaction affecting late-life depression among older adults: A nationwide,

longitudinal survey in Taiwan. Archives of Gerontology and Geriatrics, 50, S34–

S38. doi:10.1016/s0167-4943(10)70010-8

Lunenfeld, B., & Stratton, P. (2013). The clinical consequences of an ageing world

and preventive strategies. Best Practice & Research Clinical Obstetrics &

Gynaecology, 27(5), 643–659. doi: 10.1016/j.bpobgyn.2013.02.005

Lewis, S. E. (1995). The Social Construction of Depression : Experience , Discourse

and Subjectivity . VOLUME 1, 1(July).

Livingston, G., Hawkins, A., Graham, N., & Blizard, B. O. B. (1990). The Gospel

Oak Study : prevalence rates of dementia , depression and activity limitation

among elderly residents in Inner London, 137–146.

Lue, B., Chen, L., & Wu, S. (2010). Health , financial stresses , and life satisfaction

affecting late-life depression among older adults : a nationwide , longitudinal

survey in Taiwan. DART***, 50, S34–S38. http://doi.org/10.1016/S0167-

4943(10)70010-8

97 | P a g e

Maguen, S., Luxton, D. D., Skopp, N. A., & Madden, E. (2012). Gender differences in

traumatic experiences and mental health in active duty soldiers redeployed

from Iraq and Afghanistan. Journal of Psychiatric Research, 46(3), 311–316. doi:

10.1016/j.jpsychires.2011.11.007

Mallon, L., Broman, J.-E., & Hetta, J. (2000). Relationship between insomnia,

depression, and mortality: A 12-Year follow-up of older adults in the community.

International Psychogeriatrics, 12(3), 295–306.

doi:10.1017/s1041610200006414

Manber, R., & Chambers, A. S. (2009). Insomnia and depression: A multifaceted

interplay. Current Psychiatry Reports, 11(6), 437–442. doi:10.1007/s11920-009-

0066-1

Markides, K. S., & Farrell, J. (1985). Marital status and depression among Mexican

Americans. Social Psychiatry, 20(2), 86–91. doi:10.1007/bf00594985

Marmot, M. (2003). Health, wealth and lifestyles of the older population in England:

The 2002 English longitudinal study of ageing. London: IFS.

Mayer, S. E., & Jencks, C. (1989). Poverty and the distribution of material hardship.

The Journal of Human Resources, 24(1), 88. doi:10.2307/145934

MCNAMARA, D. (2006). Anxiety and sleep problems predict depression in elderly.

Internal Medicine News, 39(19), 27. doi:10.1016/s1097-8690(06)74282-2

Mendes De Leon, C. F., Rapp, S. S., & Kasl, S. V. (1994). Financial strain and

symptoms of depression in a community sample of elderly men and women: A

longitudinal study. Journal of Aging and Health, 6(4), 448–468.

doi:10.1177/089826439400600402

Mendenhall, E., Narayanan, G., & Prabhakaran, D. (2012). Short Report :

Complications Depression and diabetes in India : perspectives and

recommendations, (Ccdc), 308–311. http://doi.org/10.1111/j.1464-

5491.2012.03708.x

Millard, P. H. (1983). BRITISH, 287(6389), 375–376.

Millán-Calenti, J. C., Sánchez, A., Lorenzo, T., & Maseda, A. (2011). Depressive

symptoms and other factors associated with poor self-rated health in the elderly:

98 | P a g e

Gender differences. Geriatrics & Gerontology International, 12(2), 198–206.

doi:10.1111/j.1447-0594.2011. 00745.x

Mirowsky, J., & Ross, C. E. (1999). Economic hardship across the life course.

American Sociological Review, 64(4), 548. doi:10.2307/2657255

Mirowsky, J., & Ross, C. E. (2001). Age and the effect of economic hardship on

depression. Journal of Health and Social Behavior, 42(2), 132.

doi:10.2307/3090174

Murphy, J. G., Yurasek, A. M., Dennhardt, A. A., Skidmore, J. R., McDevitt-Murphy,

M. E., MacKillop, J., & Martens, M. P. (2013). Symptoms of depression and PTSD

are associated with elevated alcohol demand. Drug and Alcohol Dependence,

127(1-3), 129–136. doi: 10.1016/j.drugalcdep.2012.06.022

Murray, J., Banerjee, S., Byng, R., Tylee, A., Bhugra, D., & Macdonald, A. (2006).

Primary care professionals’ perceptions of depression in older people: A

qualitative study. Social Science & Medicine, 63(5), 1363–1373. doi:

10.1016/j.socscimed.2006.03.037

Nikolic, M. (2015). Prevalence of comorbid depression and obesity in general

practice. British Journal of General Practice, 65(638), 451–451.

doi:10.3399/bjgp15x686449

Nolan, B. (2009). Sleep disturbance and depression recurrence in community-

dwelling older adults: A prospective study. Yearbook of Neurology and

Neurosurgery, 2009, 171–172. doi:10.1016/s0513-5117(09)79014-3

Nolen-Hoeksema, S. (1987). Sex differences in unipolar depression: Evidence and

theory. Psychological Bulletin, 101(2), 259–282. doi:10.1037/0033-

2909.101.2.259

Orhan, F. Ö., Tuncel, D., Taş, F., Demirci, N., Özer, A., & Karaaslan, M. F. (2011).

Relationship between sleep quality and depression among elderly nursing home

residents in turkey. Sleep and Breathing, 16(4), 1059–1067. doi:10.1007/s11325-

011-0601-2

Offici, R. (2001). WORLD HEALTH REPORT 200t-MENTAL HEALTH, 4(July).

99 | P a g e

Pallesen, S., Nordhus, I. H., Kvale, G., Havik, O. E., Nielsen, G. H., Johnsen, B. H., …

Hjeltnes, L. (2002). Psychological characteristics of elderly insomniacs. Scandinavian

Journal of Psychology, 43(5), 425–432. doi:10.1111/1467-9450.00311

Patten, S. B. (2001). Long-term medical conditions and major depression in a

Canadian population study at waves 1 and 2. Journal of Affective Disorders,

63(1-3), 35–41. doi:10.1016/s0165-0327(00)00186-5

Perreira, K. M., & Sloan, F. A. (2002). Excess alcohol consumption and health

outcomes: A 6-year follow-up of men over age 50 from the health and

retirement study. Addiction, 97(3), 301–310. doi:10.1046/j.1360-0443.2002.

00067.x

Potts, M. K. (1997). Social support and depression among older adults living alone:

The importance of friends within and outside of a retirement community. Social

Work, 42(4), 348–362. doi:10.1093/sw/42.4.348

Price, R. H., Choi, J. N., & Vinokur, A. D. (2002). Links in the chain of adversity

following job loss: How financial strain and loss of personal control lead to

depression, impaired functioning, and poor health. Journal of Occupational

Health Psychology, 7(4), 302–312. doi:10.1037/1076-8998.7.4.302

Pudrovska, T. (2005). The sense of mastery as a mediator and moderator in the

association between economic hardship and health in late life. Journal of Aging

and Health, 17(5), 634–660. doi:10.1177/0898264305279874

QUINLAN, J. R. (1999). Simplifying decision trees. International Journal of Human-

Computer Studies, 51(2), 497–510. doi:10.1006/ijhc.1987.0321

Regan, C. O., Kearney, P. M., Savva, G. M., Cronin, H., & Kenny, R. A. (2013). Age and

sex differences in prevalence and clinical correlates of depression: First results

from the Irish longitudinal study on Ageing. International Journal of Geriatric

Psychiatry, 28(12), 1280–1287. doi:10.1002/gps.3955

Rice, N. E., Lang, I. A., Henley, W., & Melzer, D. (2010). Common health predictors of

early retirement: Findings from the English longitudinal study of Ageing. Age

and Ageing, 40(1), 54–61. doi:10.1093/ageing/afq153

Rifel, J., Švab, I., Rotar Pavlič, D., King, M., & Nazareth, I. (2010). Longstanding

disease, disability or infirmity and depression in primary care. Wiener klinische

Wochenschrift, 122(19-20), 567–571. doi:10.1007/s00508-010-1463-5

100 | P a g e

Roberts, R. E., Shema, S. J., Kaplan, G. A., & Strawbridge, W. J. (2000). Sleep

complaints and depression in an aging cohort: A prospective perspective.

American Journal of Psychiatry, 157(1), 81–88. doi:10.1176/ajp.157.1.81

Rodgers, B., Korten, A. E., Jorm, A. F., Jacomb, P. A., Christensen, H., & Henderson, A.

S. (2000). Non-linear relationships in associations of depression and anxiety with

alcohol use. Psychological Medicine, 30(2), 421–432.

doi:10.1017/s0033291799001865

Rosenvinge, H. P. (1988). Points: Late life depression: Undertreated? BMJ,

296(6631), 1263–1263. doi:10.1136/bmj.296.6631.1263-c

Russell, D. W., & Cutrona, C. E. (1991). Social support, stress, and depressive

symptoms among the elderly: Test of a process model. Psychology and Aging,

6(2), 190–201. doi:10.1037/0882-7974.6.2.190

Ross, C. E. (2016). Age and Depression Author ( s ): John Mirowsky and Catherine

E . Ross Source : Journal of Health and Social Behavior , Vol . 33 , No . 3

( Sep ., 1992 ), pp . 187-205 Published by : American Sociological Association

Stable URL : http://www.jstor.org/stable/2137349 Accessed : 01-04-2016 12 :

35 UTC Your use of the JSTOR archive indicates your acceptance of the Terms

& Conditions of Use , available at, 33(3), 187–205.

Sadler, W. A., & Weiss, R. S. (1975). Loneliness: The experience of emotional and

social isolation. Contemporary Sociology, 4(2), 171. doi:10.2307/2062224

Santini, Z. I., Koyanagi, A., Tyrovolas, S., & Haro, J. M. (2015). The association of

relationship quality and social networks with depression, anxiety, and suicidal

ideation among older married adults: Findings from a cross-sectional analysis

of the Irish longitudinal study on Ageing (TILDA). Journal of Affective Disorders,

179, 134–141. doi: 10.1016/j.jad.2015.03.015

Sargent-Cox, K., Butterworth, P., & Anstey, K. J. (2011). The global financial crisis

and psychological health in a sample of Australian older adults: A longitudinal

study. Social Science & Medicine, 73(7), 1105–1112. doi:

10.1016/j.socscimed.2011.06.063

Saunders, P. A., Copeland, J. R., Dewey, M. E., Davidson, I. A., McWilliam, C., Sharma,

V., & Sullivan, C. (1991). Heavy drinking as a risk factor for depression and

101 | P a g e

dementia in elderly men. Findings from the Liverpool longitudinal community

study. The British Journal of Psychiatry, 159(2), 213–216.

doi:10.1192/bjp.159.2.213

Schofield, D. J., Shrestha, R. N., Percival, R., Kelly, S. J., Passey, M. E., & Callander, E.

J. (2011). Quantifying the effect of early retirement on the wealth of individuals

with depression or other mental illness. The British Journal of Psychiatry, 198(2),

123–128. doi:10.1192/bjp.bp.110.081679

Scott, J. (1989). Can depression be prevented? Psychiatric Bulletin, 13(10), 583–583.

doi:10.1192/pb.13.10.583

Scott, D. (2007). Resolving the quantitative–qualitative dilemma: A critical realist

approach. International Journal of Research & Method in Education, 30(1), 3–17.

doi:10.1080/17437270701207694

Seeman, T. E. (2000). Health promoting effects of friends and family on health

outcomes in older adults. American Journal of Health Promotion, 14(6), 362–370.

doi:10.4278/0890-1171-14.6.362

Singh, A., & Misra, N. (2009). Loneliness, depression and sociability in old age.

Industrial Psychiatry Journal, 18(1), 51. doi:10.4103/0972-6748.57861

Stack, S., & Eshleman, J. R. (1998). Marital status and happiness: A 17-Nation study.

Journal of Marriage and the Family, 60(2), 527. doi:10.2307/353867

Sukegawa, T., Itoga, M., Seno, H., Miura, S., Inagaki, T., Saito, W., … Horiguchi, J.

(2003). Sleep disturbances and depression in the elderly in Japan. Psychiatry and

Clinical Neurosciences, 57(3), 265–270. doi:10.1046/j.1440-1819.2003. 01115.x

Sun, W., Watanabe, M., Tanimoto, Y., Shibutani, T., Kono, R., Saito, M., … Kono, K.

(2007). Factors associated with good self-rated health of non-disabled elderly

living alone in Japan: A cross-sectional study. BMC Public Health, 7(1), .

doi:10.1186/1471-2458-7-297

Supplemental material for pain, depression, and fatigue: Loneliness as a longitudinal

risk factor (2013). Health Psychology. doi: 10.1037/a0034012.supp

Son, Y., Kim, H., Kim, E., Choi, S., & Candidate, D. (2010). Application of Support

Vector Machine for Predic- tion of Medication Adherence in Heart Failure Pa-

tients, 16(4), 253–259. http://doi.org/10.4258/hir.2010.16.4.253

102 | P a g e

Steptoe, A., Breeze, E., Banks, J., & Nazroo, J. (2013). Cohort Profile : The English

Longitudinal Study of Ageing, (November 2012), 1640–1648.

http://doi.org/10.1093/ije/dys168

Subramaniam, H., & Mitchell, A. J. (2005). Reviews and Overviews Prognosis of

Depression in Old Age Compared to Middle Age : A Systematic Review of

Comparative Studies, (September), 1588–1601.

Taylor, D. J., Lichstein, K. L., Durrence, H. H., Reidel, B. W., & Bush, A. J. (n.d.).

Epidemiology of Insomnia , Depression , and Anxiety.

Trief, P. M. (2007). Depression in elderly diabetes patients. Diabetes Spectrum,

20(2), 71–75. doi:10.2337/diaspect.20.2.71

Van Beljouw, I. M. J., van Exel, E., de Jong Gierveld, J., Comijs, H. C., Heerings, M.,

Stek, M. L., & van Marwijk, H. W. J. (2014). “Being all alone makes me sad”:

Loneliness in older adults with depressive symptoms. International

Psychogeriatrics. doi:10.1017/s1041610214000581

Van de Velde, S., Bracke, P., & Levecque, K. (2010). Gender differences in

depression in 23 European countries. Cross-national variation in the gender gap

in depression. Social Science & Medicine, 71(2), 305–313. doi:

10.1016/j.socscimed.2010.03.035

Van den Brink, R. H. S., Ormel, J., Tiemens, B. G., Smit, A., Jenner, J. A., van der Meer,

K., & van Os, T. W. D. P. (2002). Predictability of the one-year course of

depression and generalized anxiety in primary care. General Hospital Psychiatry,

24(3), 156–163. doi:10.1016/s0163-8343(02)00183-4

Van Gool, C. H. (2003). Relationship between changes in depressive symptoms and

unhealthy lifestyles in late middle aged and older persons: Results from the

longitudinal aging study Amsterdam. Age and Ageing, 32(1), 81–87.

doi:10.1093/ageing/32.1.81

Van’t Veer-Tazelaar, P. J. (n., van Marwijk, H. W. J., Jansen, A. P. D. (d., Rijmen, F.,

Kostense, P. J., van Oppen, P., … Beekman, A. T. F. (2008). Depression in old age

(75+), the PIKO study. Journal of Affective Disorders, 106(3), 295–299. doi:

10.1016/j.jad.2007.07.004

103 | P a g e

Vikström, J., Bladh, M., Hammar, M., Marcusson, J., Wressle, E., & Sydsjö, G. (2011).

The influences of childlessness on the psychological well-being and social

network of the oldest old. BMC Geriatrics, 11(1),. doi:10.1186/1471-2318-11-78

Vafaei, A., Ahmed, T., Falc, N., & Zunzunegui, M. V. (2016). Depression , Sex and

Gender Roles in Older Adult Populations : The International Mobility in Aging

Study ( IMIAS ), 72, 1–15. http://doi.org/10.1371/journal.pone.0146867

Vanessa, L., Banerjee, S., Bhugra, D., Kuljeet, S., Turner, S., & Joanna, M. (2006).

Coping with depression in later life : a qualitative study of help-seeking in three

ethnic groups, (July), 1375–1383. http://doi.org/10.1017/S0033291706008117

Verplancke, T., Looy, S. Van, Benoit, D., Vansteelandt, S., Depuydt, P., &

Decruyenaere, J. (2008). BMC Medical Informatics and Support vector machine

versus logistic regression modeling for prediction of hospital mortality in

critically ill patients with haematological malignancies, 8, 1–8.

http://doi.org/10.1186/1472-6947-8-56

Vitral, R. W. F., Campos, M. J. S., & Fraga, M. R. (2013). The null hypothesis.

American Journal of Orthodontics and Dentofacial Orthopedics, 144(4), 498–

499. http://doi.org/10.1016/j.ajodo.2013.08.010

Wagner, D. C., & Short, J. L. (2014). Longitudinal predictors of self-rated health and

mortality in older adults. Preventing Chronic Disease, 11, doi:10.5888/pcd11.130241

Weeks, D. G., Michela, J. L., Peplau, L. A., & Bragg, M. E. (1980). Relation between

loneliness and depression: A structural equation analysis. Journal of Personality

and Social Psychology, 39(6), 1238–1244. doi:10.1037/h0077709

Weissman, M. M. (1977). Sex differences and the Epidemiology of depression.

Archives of General Psychiatry, 34(1), 98.

doi:10.1001/archpsyc.1977.01770130100011

Weissman, M. M., & Klerman, G. L. (1985). Gender and depression, 416–420

Weyerer, S., Eifflaender-Gorfer, S., Wiese, B., Luppa, M., Pentzek, M., Bickel, H., …

Riedel-Heller, S. G. (2013). Incidence and predictors of depression in non-

demented primary care attenders aged 75 years and older: Results from a 3-

year follow-up study. Age and Ageing, 42(2), 173–180.

doi:10.1093/ageing/afs184

104 | P a g e

Whelan, C. T. (1993). The role of social support in mediating the psychological

consequences of economic stress. Sociology of Health and Illness, 15(1), 86–101.

doi:10.1111/1467-9566.ep11343797

Wild, B., Herzog, W., Schellberg, D., Lechner, S., Niehoff, D., Brenner, H., … Raum, E.

(2011). Association between the prevalence of depression and age in a large

representative German sample of people aged 53 to 80 years. International

Journal of Geriatric Psychiatry. doi:10.1002/gps.2728

Williams, M. M. (2006). Treating depression to prevent diabetes and its

complications: Understanding depression as a medical risk factor. Clinical

Diabetes, 24(2), 79–86. doi:10.2337/diaclin.24.2.79

Wong, N. M. L., Liu, H.., Lin, C., Huang, C.., Wai, Y.., Lee, S.., & Lee, T. M. C. (2016).

Loneliness in late-life depression: Structural and functional connectivity during

affective processing. Psychological Medicine, 46(12), 2485–2499.

doi:10.1017/s0033291716001033

Yang, Y. (2007). Is old age depressing? Growth Trajectories and cohort variations in

late-life depression. Journal of Health and Social Behavior, 48(1), 16–32.

doi:10.1177/002214650704800102

Zarit, S. H., Femia, E. E., Gatz, M., & Johansson, B. (1999). Prevalence, incidence and

correlates of depression in the oldest old: The OCTO study. Aging & Mental Health,

3(2), 119–128. doi:10.1080/13607869956271

Zimmerman, F. J., & Katon, W. (2005). Socioeconomic status, depression disparities,

and financial strain: What lies behind the income-depression relationship?

Health Economics, 14(12), 1197–1215. doi:10.1002/hec.1011

105 | P a g e

Appendix

Appendix A (Questions/measures/responses/re-coded variables)

Questions asked in the ELSA survey for the factors considered in this study,

with their response options.

The name and question related to each of the ten risk factors considered in this study

and their recoded name (if it was required) are summarised as below: -

1. Variable Name: IndSex -> Recoded -> Age

Question asked at the time of interview: Respondent’s Sex

1. Male

2. Female

Risk factor considered from this variable: Gender

2. Variable Name: IndAgeR

Question asked at the time of interview: IndAgeR is derived from variable -

INDOB and variable - INTDAT. The current age of the respondent was

computed from the date of birth of the respondent (INDOB) to the date of

interview (INTDAT).

Risk factor considered from this variable: Age

3. Variable Name: DiMar -> Recoded -> Marital Status

Question asked at the time of interview: What is your current legal marital

status?

1. Single that is never married

2.Married, first and only marriage

3. Remarried, second or later marriage

4.Legally separated

5.Divorced

6.Widowed

Risk factor considered from this variable: Marital Status

106 | P a g e

4. Variable Name: SCCHD

Question asked at the time of interview: Do you have any children?

1. Yes

2. No

Risk factor considered from this variable: Children Availability

5. Variable Name: PScedC

Question asked at the time of interview: Much of the time during the past

week, your sleep was restless?

1 Yes

2 No

Risk factor considered from this variable: Insomnia

6. Variable Name:Hehelf

Question asked at the time of interview: Would you say your health is...

1 Excellent,

2 very good,

3 good,

4 fair,

5 poor

Risk factor considered from this variable: Self rated health

7. Variable Name: Heill

Question asked at the time of interview: Do you have any long-standing

illness, disability or infirmity? By longstanding I mean anything that has

troubled you over a period of time, or

that is likely to affect you over a period of time?

1 Yes

2 No

107 | P a g e

Risk factor considered from this variable: Long standing

illness/disability/infirmity

8. Variable Name: PScedE

Question asked at the time of interview: Much of the time during the past

week, you felt lonely?

1 Yes

2 No

Risk factor considered from this variable: Loneliness

9. Variable Name: SCQOLI

Question asked at the time of interview: Shortage of money stops me from

doing things I want to do

1 Often

2 Sometimes

3 Not often

4 Never

Risk factor considered from this variable: Financial Strain

10. Variable Name: Heala

Question asked at the time of interview: In the past 12 months have you taken

an alcoholic drink ...

1 twice a day or more,

2 daily or almost daily,

3 once or twice a week,

4 once or twice a month,

5 special occasions only,

6 or, not at all?

Risk factor considered from this variable: Alcohol consumption

108 | P a g e

11. Dependent Variable Name : PScedA

Question asked at the time of interview: Much of the time during the past

week, you felt depressed?

1 Yes

2 No

Risk factor considered from this variable: Depression

Variables that were recoded

The variable for age was recoded to make it as a categorical variable with age group

categories as: “50-59”; “60-69”; “70-79”; and “80+”. Few values less than 50 were

present which were changed to missing values. The variable for marital status was

also recoded to include categories that were more meaningful and consistent

throughout all waves as first two waves of data didn’t have civil partner categories

that were present from wave 3, but they had very few numbers of respondents in that,

therefore civil partner categories were combined with legally separated and divorced

categories. Also, legally separated and divorced categories were also combined

together to form one category as one of the category had very low frequency

throughout all waves. Therefore, the final categories of marital status were: “Single,

that is never married”; “Married, first and only marriage”; “Remarried, second or

later marriage”; and “Legally separated/divorced/civil partner” and “Widowed”.

Alcohol variable was present as “HEALA” for (Wave 1) and “SCAKO” for (Wave 2

to 6). Wave 1 had 5 response categories. And Wave 2 to 6 had 8 categories, which

were finally combined to form five categories in the same order as present in wave 1

as: “Almost daily”; “Once or twice a week”; “Once or twice a month”; “Special

occasion only”; “Not at all”. The variable for depression was recoded from the

categories “1: Depressed” and “2: Not depressed” to “0: Not depressed” and “1:

Depressed” for the purposes of this research.

109 | P a g e

Missing values

For each of the variable the response categories as “Not applicable” (-1), “Don’t

know” (-8), “Refusal” (-9), and “Schedule not applicable” (-2) were changed to

missing values in SPSS before carrying out any analysis. Further few variables were

re-coded for the purpose of this study which is detailed in appendix A.

Appendix B (Weka outputs)

SVM

J48- Pruned

110 | P a g e

The most relevant and predictive section selected by unpruned J48-decision tree

111 | P a g e

J48- Unpruned

Appendix C (Additional useful information)

Quantitative research method

s

Figure 1.1 The research process (Bryman & Cramer, 2001, p.3)

112 | P a g e

This whole process can be seen in the ELSA survey whose collected data have been

used in this study. Data were collected by ELSA through correlational design as the a

im was to measure changes and relationships between variables. Participants of age 5

0 and over were selected. Questionnaires related to 13 modules and a self-completion

form were made available to each participant. Data were collected for every two yea

rs. This data of different waves has been used in this study to analyse and present fin

dings related to the predictors of depression.

Chi-squared test interpretation

Continuity correction was used to interpret results of the chi-squared test for

variables - Gender, Children, Insomnia, Long standing health problem and

Loneliness as each these variable had 2 categories and 2X2 table was produced for

each of them. Pearson Chi-Square was used to interpret result of the chi-squared test

for the variable - Marital Status. And, Linear-by-Linear Association was used in case

of the variables - Age, Financial Strain and Alcohol consumption as each of them

were ordinal variables and had more than 3 categories. Further, ELSA dataset was

huge, thus, problems related to each category’s expected count were not faced.

Appendix D (Tree diagram and status of each factor)

Tree diagram

The process that was followed in SPSS to observe change in depression status across

each wave of ELSA to create a tree diagram

SORT CASES BY Depression.

SPLIT FILE SEPARATE BY Depression.

FREQUENCIES VARIABLES=Depression_2

/ORDER=ANALYSIS.

From this, change in depression status for all of the respondents from wave 1 to wave

2 was obtained

SORT CASES BY Depression. Depression_2

SPLIT FILE SEPARATE BY Depression.

113 | P a g e

FREQUENCIES VARIABLES=Depression_3

/ORDER=ANALYSIS.

Similarly, from wave 2 to wave 3

SORT CASES BY Depression Depression_2 Depression_3.

SPLIT FILE SEPARATE BY Depression Depression_2 Depression_3.

FREQUENCIES VARIABLES=Depression_4

/ORDER=ANALYSIS.

SORT CASES BY Depression Depression_2 Depression_3 Depression_4.

SPLIT FILE SEPARATE BY Depression Depression_2 Depression_3 Depression_4.

FREQUENCIES VARIABLES=Depression_5

/ORDER=ANALYSIS.

From wave 3 to wave 4

SORT CASES BY Depression Depression_2 Depression_3 Depression_4.

SPLIT FILE SEPARATE BY Depression Depression_2 Depression_3 Depression_4.

FREQUENCIES VARIABLES=Depression_5

/ORDER=ANALYSIS.

From wave 4 to wave 5

SORT CASES BY Depression Depression_2 Depression_3 Depression_4

Depression_5

SPLIT FILE SEPARATE BY Depression Depression_2 Depression_3 Depression_4

Depression_5

FREQUENCIES VARIABLES=Depression_6

/ORDER=ANALYSIS.

Finally, from wave 5 to 6, which identified the respondents who were depressed or

not depressed throughout all 6 waves.

How data of depressed and not depressed older adults was obtained

New variable depression_all was created

1) It was given a value 0 if (depression_wave1=0 & depression_wave1=0 &

depression_wave1=0 & depression_wave1=0 & depression_wave1=0 &

depression_wave1=0) -> this gave data for the respondents who were not depressed

throughout all 6 waves of ELSA.

114 | P a g e

2) It was given a value 1 if (depression_wave1=1 & depression_wave1=1 &

depression_wave1=1 & depression_wave1=1 & depression_wave1=1 &

depression_wave1=1) -> this gave data for the respondents who were depressed

throughout all 6 waves of ELSA.

Status for each of the factor

Data obtained for the respondents who were depressed or not depressed throughout

were further filter to obtain data only for depressed respondents. From which status

of each factor at the beginning and at the end of the ELSA survey was determined.

118 | P a g e

Access to Dissertation

A Dissertation submitted to the University may be held by the Department (or School) within

which the Dissertation was undertaken and made available for borrowing or consultation in

accordance with University Regulations.

Requests for the loan of dissertations may be received from libraries in the UK and overseas.

The Department may also receive requests from other organisations, as well as individuals.

The conservation of the original dissertation is better assured if the Department and/or

Library can fulfill such requests by sending a copy. The Department may also make your

dissertation available via its web pages.

In certain cases where confidentiality of information is concerned, if either the author or the

supervisor so requests, the Department will withhold the dissertation from loan or

consultation for the period specified below. Where no such restriction is in force, the

Department may also deposit the Dissertation in the University of Sheffield Library.

To be completed by the Author – Select (a) or (b) by placing a tick in the

appropriate box

If you are willing to give permission for the Information School to make your dissertation

available in these ways, please complete the following:

\

/

(a) Subject to the General Regulation on Intellectual Property, I, the author, agree to

this dissertation being made immediately available through the Department and/or

University Library for consultation, and for the Department and/or Library to

reproduce this dissertation in whole or part in order to supply single copies for the

purpose of research or private study

(b) Subject to the General Regulation on Intellectual Property, I, the author, request

that this dissertation be withheld from loan, consultation or reproduction for a

period of [ ] years from the date of its submission. Subsequent to this period, I

agree to this dissertation being made available through the Department and/or

University Library for consultation, and for the Department and/or Library to

reproduce this dissertation in whole or part in order to supply single copies for the

purpose of research or private study

Name Pranshu Bhasin

Departme Information School Date: 30/08/2016

119 | P a g e

Signed Pranshu Bhasin

To be completed by the Supervisor – Select (a) or (b) by placing a tick in the

appropriate box

(a) I, the supervisor, agree to this dissertation being made immediately available through

the Department and/or University Library for loan or consultation, subject to any

special restrictions (*) agreed with external organisations as part of a collaborative

project.

*Special

restrictio

ns

(b) I, the supervisor, request that this dissertation be withheld from loan, consultation or

reproduction for a period of [ ] years from the date of its submission. Subsequent

to this period, I, agree to this dissertation being made available through the

Department and/or University Library for loan or consultation, subject to any special

restrictions (*) agreed with external organisations as part of a collaborative project

Name

Departme

nt

Signed Date

THIS SHEET MUST BE SUBMITTED WITH DISSERTATIONS BY DEPARTMENTAL

REQUIREMENTS