my first empirical paper

Upload: sazidh1

Post on 05-Apr-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/2/2019 My First Empirical Paper

    1/23

    Page 1 of23

    Does Being in a Relationship Increase Mens Earnings? A

    Cross-sectional Study on DU Male Students

    1. Introduction:

    This study was motivated by the concept of male marriage premium, a long-standing

    question of interest in labor economics. It has been seen in many empirical studies that

    married men, on average, earn more than their single counterparts holding all else equal.

    More interestingly, cohabiting men have also been found to have more income than their

    single counterparts. In this path we found it the obvious next question: Does being in a

    relationship increase mens earnings?

    This question did not get attention in the existing literature yet and probably will not get

    attention ever. But it does not mean that it is an unimportant question, particularly in the

    context of Bangladesh. Traditionally Bangladesh is a conservative country in the case of free

    mixing between men and women; cohabiting is not allowed and being in a relationship is

    usually a much more a serious matter, in responsibility and commitment, in comparison with

    countries with advanced economy where almost all the marriage premium related studies

    took place. Unlike those countries, in a relationship is not typically a light relationship; it

    involves a greater amount of sharing and caring; it is usually associated with commitment or

    at least willingness to take the relationship to marriage. So, Bangladeshi men who are in a

    relationship are likely to have some unobserved characteristics that are considered as

    potential reasons behind marriage premium. For example, they are likely to be more stable,

    matured and forward looking; they are likely to be interested in maintaining a regular stream

    of income and thus be more tolerable to unfavorable working conditions; they may get career

  • 8/2/2019 My First Empirical Paper

    2/23

    Page 2 of23

    advice and moral support. However, some other potential reasons behind marriage premium,

    such as greater opportunity for married men to concentrate more on outside home earning

    activities due to less tasks at home, employers discrimination etc. are not applicable in the

    case of relationship. Again, relationship is comparatively short-lived than marriage and

    thus probably has less strong effect, if any. So, whether being in a relationship helps men to

    earn more than their similar single counterparts is not clear in advance and can be a question

    of empirical interest. Where formal studies relevant with marriage premium has stopped in

    cohabiting men, we think we should extend it to men who are in a relationship at least in the

    context of Bangladesh.

    In this study, using primary data of 286 male students of University of Dhaka, first we

    assessed whether students who are in a relationship have higher tendency to participate in

    income earning activities. Then we focused on only those students who have own sources of

    earning and sought if being in a relationship has any positive impact on their income.

    We chose University of Dhaka for three reasons. Firstly, it is a large university with students

    from different socioeconomic backgrounds. So we are likely to get enough variability in the

    data. Secondly, this university has students from every corners of the country and thus this

    population represents the university level students of Bangladesh better than any other

    institution. The third reason was the convenience of data collection.

    Rest of the paper is organized in the following manner: section 2 deals with the conceptual

    framework of this study, section 3 provides a brief literature review, section 4 discusses the

    empirical methods used, section 5 discusses the data, section 6 analyses the results, section 7

    admits some limitations of this study and provides some guidance for further research and

    finally, section 8 draws conclusions.

  • 8/2/2019 My First Empirical Paper

    3/23

    Page 3 of23

    2. Conceptual Framework:

    In this study we applied the concept of marriage premium in the pursuit of our objective.

    There are two broad explanations behind marriage premium. One explanation says this

    premium is a causal effect of marriage. The other explanation deals with the reverse causality

    and attributes this premium to the issues of selectivity in marriage.

    Casual effect of marriage on income works through at least five channels. Firstly, Married

    men have to perform less on household works, as traditionally their wives do most of them,

    and thus can concentrate on earning activities. This specialization allows them to earn more.

    Since being in relationship does not involve such division of labor, this process is not

    applicable in our context. Secondly, married men may get career advice and moral support

    from their spouses which may increase their earning. Thirdly, married men feel greater

    financial responsibility and this feeling induces them to earn more. In our context this process

    is applicable since most of the common expenditures in a relationship are carried by men.

    Fourthly, marriage develops regularity in lifestyle that induces regular working habit and

    increases productivity. Whether this process is applicable in our case is uncertain because

    relationships sometimes become volatile and usually less smooth than marriage. Fifthly,

    marriage makes men more tolerant, stable, matured and regular. Thus they might get higher

    wages as a result of employers discrimination for a given level of productivity. As

    employers do not care for whether a man is in a relationship status, such employers

    discrimination is not applicable in our case.

    On the other hand selectivity may cause marriage premium in two ways. Firstly, men who

    have high unobserved ability exhibit some characteristics like stability, patience, enthusiasm,

    industriousness etc. which are attractive to both employers and potential spouses. So, men

    with high wages or wage growth may have greater chances to get married. Secondly, same

  • 8/2/2019 My First Empirical Paper

    4/23

    Page 4 of23

    married men may prefer jobs that have higher wages and less non-monetary benefits. Both of

    these processes may hold in case of relationship.

    3. Literature Review

    Existing literature does not have any study on exactly the issue this paper deals with.

    However, there are plenty of relevant works on male marriage premium and some works on

    cohabiting men which are closely related to our context.

    Riber (2004) described the causes behind marriage premium and summarized different

    quantitative methodologies to empirically work with this issue. He also pointed out potential

    obstacles that are likely to occur in empirical studies in marriage premium.

    The works on marriage premium mainly follow on two empirical approaches: study of cross-

    sectional data and study of panel data. In forming empirical models in both types of studies,

    income or log of income is taken as the dependent variable, one or more than one binary

    variables like married, cohabiting divorced etc. are used as key explanatory variable and

    demographic variables like age, race etc. indicators of ability like education, IQ score etc are

    controlled.

    Among cross-sectional studies Bellas (1992), Blau and Beller (1988), Blackburn and

    Korenman (1994), Chun and Lee (2001), Hill (1979) and Krashinsky (2004) are worth

    mentioning. These cross-sectional studies typically have estimated a marriage premium

    between 6% and 35% (Cornaglia and Feldman 2010) However, the effect is usually

    overstated due to selection into marriage.

    To correct this bias of selection into marriage, panel data methods have increasingly been

    used in modern days. Among panel studies Bardasi and Taylor (2005), Cornwell and Rupert

    (1995), Korenman and Neumark (1991), Krashinsky (2004), Richardson (2000), Stratton

  • 8/2/2019 My First Empirical Paper

    5/23

    Page 5 of23

    (2002), Loughran and Zissimopoulos (2009), Neumark (1988), Rogers and Stratton (2005)

    are mentionable. These studies usually involve the use of fixed effect models. Most of the

    studies with panel data found a positive correlation between wage and marital status while

    some have found the effect to be indistinguishable from zero (e.g. Gray 1997). Panel studies

    generally conclude that there is some causal effect of marriage on wage but whether this

    effect is due to increased productivity or merely employers discrimination remains

    unresolved.

    Selectivity is a major concern in the study of marriage premium which has been admitted in

    numerous works. After controlling for selectivity, the remaining associations are possible

    causal effects. Waite and Gallagher (2000) cited many studies that indicate that marriage

    premium reflect more than just pre-existing differences in individuals economic abilities.

    These studies support causal effects.

    Some papers extended the study range from marriage to cohabitation (e.g. light 2004) and

    found the existence of similar premium among cohabiting men. However cohabiting mens

    premium is usually found smaller than that of married men.

    4. Econometric Models and Estimation Methods

    There are a good number of cross-sectional studies on marriage premium that used Ordinary

    Least Squares (OLS) method. Again, Two Stage Least Squares (TSLS) approaches, use of

    instrumental variables etc. are common. Some researchers have worked with longitudinal

    data and used Fixed Effects Model, Random Effects Models. While some of these advanced

    methods can reduce, if not eliminate, the problem of selectivity and reverse causality

    discussed in section 2, they involves computational complexity. Again, these methods differ

    in underlying assumptions and data requirements. There exist no cross-sectional or panel data

  • 8/2/2019 My First Empirical Paper

    6/23

    Page 6 of23

    that could serve our purpose of this study. So, a cross-sectional study with primary data

    became an automatic choice.

    In this study we dealt with two models. Firstly, we constructed a Linear Probability Model to

    roughly estimate if being in a relationship increases students chances to have own sources

    of earning. Then we used the concept of marriage premium and the insights from the existing

    literature to construct a Multiple Linear Regression Model to capture the ceteris paribus

    effect of being in a relationship, after controlling for controlling for family background

    variables (fathers education, mothers education, number of siblings, family income etc.),

    education related variables (SSC GPA, HSC GPA, university CGPA, year and a broad

    division of the subject currently studying, i.e. Science, Business), job related variables

    (experience and hours of work per week) etc. In estimating both models we used Ordinary

    Least Squares (OLS) method.

    For the first model, the population is all undergraduate and masters level male students of

    university of Dhaka. In the second model the population is only those undergraduate and

    masters level male students of university of Dhaka who have own sources of income. We

    excluded PhD and M.Phil. students since most of them are professional people, married and

    less available for surveying.

    The Linear Probability Model that we to estimate the effect of being in a relationship on the

    likeliness of having an own income source was as follows:

    ownsource = 0+ 1 relationship + 2 breakup + 3fathereduc + 4 mothereduc + 5siblings

    +6lfamilyinc + 7netallowance + 8 dhaka + 9sscgpa + 10 hscgpa +11 cgpa

    + 12year +13science + 14business + . (1)

    Here ownsource is a binary variable that takes value 1 if a student has own source of earnings

    and 0 if does not have any own earning source. As students can be stratified into always

  • 8/2/2019 My First Empirical Paper

    7/23

    Page 7 of23

    single, presently in a relationship and previously in a relationship, we introduced two

    relationship status related dummy variables. One is relationship whichtakes 1 if the student

    is presently in a relationship and 0 if otherwise. Another is breakup which takes 1 if the

    student was previously in a relationship that broke up. Other variables in this model were

    used to control for family background, whether student spent most of his life in Dhaka City,

    students education indicators etc. is the error term that represents unobserved

    characteristics that affects having own sources of earnings. (A complete description of all

    variables used in this model is available in Table-1 in the appendix.)

    Our parameter of key interest in this model is 1 whichrepresents the ceteris paribus effect of

    being in a relationship on the probability of having an own source of earnings. We

    estimated it and tested its significance against one-sided alternative using t-test:

    H0: 1 = 0

    H1: 1 > 0

    We chose one sided alternative because it is very much unlikely that being in a relationship

    will reduce students chances of having own sources of earnings.

    Then we concentrated on our main model:

    linc = 0 + 1 relationship + 2 breakup + 3fathereduc + 4 mothereduc + 5siblings

    + 6lfamilyinc + 7netallowance + 8 dhaka + 9sscgpa + 10 hscgpa + 11 cgpa

    + 12year + 13science + 14 business + 15 exper + 16weekly_hrs + u . (2a)

    Here we took linc as the dependent variable which is log of students self-earned monthly

    income. We took logarithmic form due to three reasons. Firstly, most of the papers in existing

    literature took income in logarithmic form. Secondly, taking logarithm allows us to capture

    percentage change in income which is convenient while judging the effects of explanatory

  • 8/2/2019 My First Empirical Paper

    8/23

    Page 8 of23

    variables. Thirdly, we experimented with level form which did not pass in functional form

    misspecification test. Again, since zero cannot enter into logarithmic function, this model

    naturally considers only those students who have own income as its population.

    Like the first model, we introduced two relationship status dummy variables here. One is

    relationship which takes 1 if the student is presently in a relationship and 0 if otherwise.

    Another is breakup which takes 1 if the student was previously in a relationship that broke

    up. In this model we used several control variables because existing literature indicates that

    there are many determinants of income and many of them are correlated with relationship

    status. We controlled for family background variables (fathers education, mothers

    education, number of siblings, family income etc.), education related variables (SSC GPA,

    HSC GPA, university CGPA, year and a broad division of the subject currently studying, i.e.

    Science, Business), job related variables (experience and hours of work per week) etc. and

    whether student spent most of his life in Dhaka City. u is the error term that represents

    unobserved characteristics that affects having own sources of earnings. (A complete

    description of all variables used in this model is available in Table-1 in the appendix.)

    In choosing the functional form, we chose linear model for its simplicity. In this model key

    explanatory variable relationship does not interact with other explanatory variables. So, it

    assumes that the effect of being in a relationship is uniform across all levels of other

    explanatory variable. We kept fathers education, mothers education, number of siblings,

    SSC GPA, HSC GPA, university CGPA and year in linear and level form because we

    assumed that marginal effects of these variables do not depend on their values. We also kept

    experience and weekly working hours in linear form. However, we tried with their quadratic

    terms to allow diminishing returns. But these quadratic terms did pass in F-tests and we

    became convinced not to include them because it might well happen that students experience

    and work hours are too small to have diminishing returns. We entered netallowance in level

  • 8/2/2019 My First Empirical Paper

    9/23

    Page 9 of23

    form since it can take both positive and negative values. Beside relationship and breakup we

    used three other dummy variables. We divide the subjects studied by students of University

    of Dhaka into three broad arbitrary categories: Science, Social Sciences & Humanities and

    Business. To capture the effects of these differences, we assigned two dummies: science and

    business. The rest dummy variable is dhaka which takes value 1 if the student spent major

    part of his life in Dhaka. We took family income in log form to reduce the variability. The

    coefficient of lfaminc thus represents constant elasticity while coefficients of the other

    regressors represent semi-elasticity.

    In this model, our parameter of key interest is 1, the coefficient of relationship, which

    represents the ceteris paribus effect of being in a relationship on linc. Its sample

    counterpart will give an unbiased estimator of 1 if error term u is uncorrelated with all

    explanatory variables. Unfortunately, this model lacks this property because of the omission

    of ability, aptitude, presentation skill, appearance etc. variables in this model. These variables

    have impact on students income and are likely to be correlated with relationship status.

    Ability and aptitude cannot be observed and measured. So, their obvious omission causes

    bias. One way to reduce the bias is to assign proxy variables like Intelligent Quotient (IQ),

    results of standardized tests etc. But in our case even these data were not possible to acquire.

    On the other hand appearance, presentation skill and smartness can be observed but are

    largely subjective and tough to measure. So, we had to omit them too. All of these five

    omitted variables have positive impact on earning and likely to be positively related with

    being in a relationship. So, they are likely to cause upward bias in the estimated impact of

    relationship.

    Checking for heteroskedasticty requires having a proper functional form. Our model

    specification passed Ramseys Regression Specification Error Test (RESET). After the

    RESET test, we inquired for heteroskedasticity using the White Test and found no strong

  • 8/2/2019 My First Empirical Paper

    10/23

    Page 10 of23

    evidence of heteroskedasticty. We are not sure if the Normality Assumption holds but it

    should not be a problem as our sample size is large.

    After estimating 1, we tested its significance using t-test against one-sided alternative

    hypothesis:

    H0 :1 = 0

    H1 :1 > 0

    We selected one sided alternative because it is very much unlikely that being in a

    relationship will reduce students income.

    Finally we used a slightly different specification of (2a):

    linc = 0 + 1 relationship + 2 breakup + 3fathereduc + 4 mothereduc + 5siblings

    + 6lfamilyinc + 7netallowance + 8 dhaka + 9sscgpa + 10 hscgpa + 11 cgpa

    + 12year + 13science + 14 business + 15 exper + u . (2b)

    The only difference between these two specifications is that we dropped hours of work per

    week in (2b). So,1 in model (2b) captures that effect of being in a relationship on income

    that comes through the channel of increased work hour.

    5. The Data

    All the data used in this paper is taken from a primary survey that we conducted on 1st

    and 2nd

    November 2011 at University of Dhaka. We requested students to participate in a survey of

    Economics Department and did not tell them the topic or showed the questionnaire in

    advance. We didnt pressurize anyone and only the willing students were given a

    questionnaire that did not ask any identification information. The questionnaire sought

    information about their income, relationship status, family income, parents education,

  • 8/2/2019 My First Empirical Paper

    11/23

    Page 11 of23

    number of siblings, background, present and previous academic records, whether they receive

    allowance or have to support their family and if, then by how much amount etc. (The survey

    questionnaire is available in the appendix.)

    We surveyed undergrad and masters level male students. PhD and M.Phil. students were

    excluded from the survey since they are mostly professional and married people. The survey

    took place at all major academic buildings, some randomly chosen student halls, TSC,

    playgrounds, Central Library, and university bus stoppages. In case of student dormitories,

    we first randomly picked 4 halls from all 13 male dormitories, then collected associated room

    numbers and picked some of the rooms randomly. We surveyed after 11 p.m. so that most of

    the resident students are available in their rooms. We attempted to survey all students who

    were present at the chosen rooms and around 80% of them agreed to participate in the survey

    and the rest rejected. Unfortunately, we could not conduct the survey in a purely random

    manner in other places. While surveying in academic buildings, we went to some random

    stories and targeted students standing in the corridor or waiting for the class to start. In this

    case response rate was around 60%. While surveying at the Central Library, we went to each

    floor, chose closest and furthest tables from the main doors and then approached to the people

    sitting at the corners of the tables. Here we got almost 90% responses. In case of TSC,

    playgrounds and bus stoppages we targeted people sitting alone or in groups. Around 80%

    students participated in this case. Finally total response reached to 287 but we found one

    questionnaire completely blank. So, our data set includes response of 286 students. However,

    not everyone answered all questions. Again, as the actual proportions in population of three

    stratums: always single, previously in a relationship and presently in a relationship are

    unknown, we could not maintain them in the sample.

    In the data, around 56% students have own source of earning. Length of having own source

    of earning, denoted by exper, has a mean of approximately 13 months while the maximum

  • 8/2/2019 My First Empirical Paper

    12/23

    Page 12 of23

    length is as high as 84 months. Students average income per month is Tk. 3364 while the

    maximum income is Tk. 35000. Data shows the students on average works for 8 hours per

    week. However, this average includes people who do not have own source of earning. If we

    consider only those who has own source of earning then the average working hour per week

    reaches to 15. Among 286 students, 278 revealed their relationship status. Around 60%

    reported that they are single and were never in relationship, 14% said they are presently in

    relationship and the rest said they are now single but previously were in relationship. Only

    around 50% students who are presently in relationship reported their spouses income which

    is only 10% of the total sample size. So, we had to exclude this variable from our models

    even though it was important. Around 92% students whom we surveyed reported their family

    income the average family income was around Tk. 23,000 per month while the minimum and

    the maximum are Tk. 0 and Tk. 150,000 per month respectively. Around 90% students

    reported their parents education. In case of fathers education the average was 12 years and

    in case of mothers education the average was 9 years. In these cases standard deviations

    were around 4 years. Around 32% students in the sample spent major portion of their life up

    to HSC in Dhaka city. The sample consists of 23% science students, 32% business students

    and the rest from Social Sciences or Humanities. Students in the sample on average get

    allowance of Tk. 2,350 per month. A complete summary of the data is given in Table-2 in the

    appendix.

    Our inspections found no evidence of endogenous sample selection while there was a little

    exogenous sample selection in case of parents education: those around 10% students who

    did not report fathers education and those around 10% students who did not report mothers

    education are likely to have parents with less education. However, these endogenous sample

    selections are harmless and do not cause any harm in regression results.

  • 8/2/2019 My First Empirical Paper

    13/23

    Page 13 of23

    6. Results

    Regression results are given in Table-3 in the appendix. Since we did not find evidence of

    heteroskedasticty in White Test, actual standard errors have been reported. As we did not find

    exclusion of any apparently influential observation to alter regression results significantly, we

    did not drop any observation as outlier. Please note that number of observations in models

    (2a) and (2b) is remarkably fewer than that of model (1) because the previous two models

    treat only those students who have own income as their populations as described in section 3.

    In model (1), estimated coefficient ofrelationship is 0.208. It means that students who are in

    a relationship, have higher probability of around 0.21 of having own sources of earning than

    their always single counterparts. This estimator is statistically significant at 1 percent level of

    significance against both one-sided and two-sided alternatives. This result definitely has

    practical significance because having own sources of earning depends on many factors and as

    a marginal effect of 0.21 on its probability is large.

    Regression results also say that students who were previously in a relationship that broke up

    have 0.14 higher probability, which is practically significant, to have own sources of earning

    than their similar always single counterparts. However, the coefficient of breakup is not

    statistically significant which implies that this effect is statistically not different from zero in

    population.

    The results say students likelihood of having own earning source is negatively related with

    fathers education and positively with mothers education after controlling for other factors. A

    4 years increase in fathers education reduces probability of having own source of earning by

    approximately 0.11 while a 4 years increase in mothers education increases the probability of

    having own source of earning by approximately 0.09. Though the effects are not very large,

    they are not too small to neglect. Both of the coefficients are statistically significant.

  • 8/2/2019 My First Empirical Paper

    14/23

    Page 14 of23

    Estimated coefficient of linc is -0.09 which means one percent increase in monthly income

    reduces the likelihood of having own sources of earning by 0.0009 which is practically very

    small, though the coefficient is statistically significant at 5 percent level. The estimated

    coefficient ofnetallowance is-0.050 which says an additional one thousand Tk. netallowance

    decreases the probability of having own earning source by 0.05. Though this effect is small

    the coefficient is statistically very significant. The estimated coefficient of dhaka is 0.005

    which means students from Dhaka background are less than one percent more likely to have

    own earning sources. However this coefficient is not statistically significant that is in the

    population there is no difference, in terms of probability of having own source of income,

    between students from Dhaka background and students from outside Dhaka. The coefficients

    of SSC and HSC GPAs say that these variables have very little positive impact on the

    likelihood of having own source of earning. These coefficients are statistically insignificant

    too. Coefficient of university CGPA is -0.126 which says a one point increase in university

    CGPA (Which is a very large change) reduces students likelihood of having own sources of

    earning by 0.126 which is a moderate decrease. Results show that Business students have

    0.108 higher probability and Science students have 0.04 lower probability of having own

    source of earning than similar Social-sciences & Humanities students.

    In case of model (2a), estimated coefficient of relationship is 0.184. It means that students

    who are in a relationship have around 18 percent higher income than their similar single

    counterparts. The estimatorrelationship is found statistically significant at 5 percent level of

    significance against one-sided alternative and at 10 percent level against two-sided

    alternative. This result definitely has economic significance because 18 percent increase in

    income is large.

    Interestingly, estimated coefficient of breakup is 0.316 that means students who were

    previously in a relationship that broke up have 32 percent higher income than their always

  • 8/2/2019 My First Empirical Paper

    15/23

    Page 15 of23

    single counterparts. So, effect ofbreakup is higher than the effect of relationship! And the

    coefficient ofbreakup is statistically significant at 1 percent level. How can we explain this

    result? Can we still consider the coefficient ofrelationship as a causal effect on income?

    We propose two plausible explanations. First explanation: people have a natural tendency not

    to reduce their income level. They either maintain a level of income or try to increase it if

    possible. It might happen that students who were previously in a relationship that broke up

    reached to a higher income stream than their always single counterparts while they were in a

    relationship. After the breakup, due to the natural tendency just stated, they might be

    unwilling to reduce their income. Rather, they might increase income using their more

    flexible time. (As we have controlled for weekly working hour in (2a), we must not attribute

    their higher income to increase working hour due to their additional free time.) Second

    explanation: People with higher level of ability, aptitude, appearance, presentation skill and

    smartness are more likely earn more. But people who have these attributes in greater extent

    are higher valued in the market for spouses. As getting a spouse is easier for them, their cost

    of breakup is relatively low. So, they are less tolerable to the frictions of a relationship and

    more likely to break up. Thus it may happen that the effect of breakup on income is not

    causal rather the breakup group simply represents people with higher degree of ability,

    aptitude, appearance, presentation skill and smartness which brings them higher income.

    While the first explanation acknowledges the causal effect of being in a relationship on

    income, the second explanation attributes the result to selectivity.

    Note that the estimated coefficient of relationship is likely to overstate the actual partial

    effect as we could not control for ability, aptitude, appearance, presentation skill and

    smartness (discussion in section 4)

  • 8/2/2019 My First Empirical Paper

    16/23

    Page 16 of23

    Let us now see the effects of the control variables. Estimated coefficient of lfaminc is 0.145

    which says a one percent increase in family income increases students income, on average,

    only by 0.15 percent. Though this coefficient is statistically significant at 5 percent level,

    practically this effect is too small to recognize. So, we can say, income of DU students is

    almost perfectly inelastic to their family income.Net allowance is found statistically very

    significant. Its estimated coefficient -0.064 says a one thousand taka increase in net

    allowance decreases income, on average, by 6.4 percent which is a moderate influence.

    Coefficient of mothers income is 0.030 and it says that a 4 years increase in mothers

    education raises students income, on average, by 12 percent after controlling for other

    factors. This coefficient is statistically significant at 10 percent level. Estimated coefficient of

    HSC GPA is 0.36 which says that a one point increase in HSC GPA (which is a large

    increase) raises income, on average, by 36 percent holding all other constant. This coefficient

    is found statistically significant in 5 percent level. On the other hand, estimated coefficient of

    SSC GPA produces a counterintuitive negative sign. The estimated coefficients say a one

    point increase in SSC GPA (which is a large increase) reduces income by a moderate 10

    percent but this estimator is found statistically insignificant. Estimated coefficient of

    university CGPA is 0.013 which says that a one point increase in CGPA (which is a very

    large increase) raises income, on average, only by 1.3 percent holding all other constant. So,

    it says university CGAP has almost no practical significance on university students earning.

    Again, this coefficient is found statistically insignificant. The reason behind this result lies in

    the sources of students earnings. As most of the students depend on tuition that does not

    count performance at university and thus we get such a result. Regression results say Science

    students and Business students, on average, earns 7.6 percent and 16.7 percent respectively

    more than similar Social-sciences & Humanities students. However, statistical significances

    of relevant estimators say this is only a sample phenomenon and there is no such evidence in

  • 8/2/2019 My First Empirical Paper

    17/23

    Page 17 of23

    the population. According to the regression results, a one month increase in experience

    increase income by 0.2 percent. This effect is practically and statistically insignificant.

    Estimated coefficient of weekly hours of work is 0.007 which says an additional hours of

    work only brings a 0.7 percent increase in income which is little. This estimator is also

    statistically insignificant.

    In specification (2b) we dropped hours of work per week to allow include impact of

    relationship through increased working hour. In this specification, estimated coefficient of

    relationship is 0.192. It means that students who are in a relationship, on average, have

    around 19 percent higher income than their similar single counterparts. Like specification

    (2a) the estimator of relationship is found statistically significant at 5 percent level of

    significance against one-sided alternative and at 10 percent level against two-sided

    alternative. So, from this specification we understand that only a very little portion of the

    effect ofrelationship comes through the channel of increased working hour.

    In case of other coefficients, estimated values and statistical significance of the estimators are

    almost same as specification (2a).

    We see we have some important variables statistically insignificant. Our calculation of

    Variance Inflating Factor says this phenomenon did not happen because of multicollinearity

    problem. Probably we have experienced such result because OLS might not be in appropriate

    in this problem. Use of Two Stages Least Squares (2SLS) approach or Panel data methods

    might solve this problem.

  • 8/2/2019 My First Empirical Paper

    18/23

    Page 18 of23

    7. Some Limitations of the Study and Guidance for Future Work:

    OLS is not a strong enough model to deal with our objective. 2SLS or fixed effect models

    would have worked better to deal with the selectivity problem. We lack data on some

    important variables (discussed in section 4). Getting data on these variables or at least there

    proxies would make results better. Besides, we admit that the data set we used lacks in

    random sampling properties to some extent (discussed in section 5).

    Further works on this topic should be caring to eliminate these drawbacks. In addition, using

    duration of relationship, duration of break period and interactions of explanatory variables

    would be interesting.

    8. Conclusion:

    This study finds that students in a relationship are around 21 percent more likely to have

    own sources of earnings than their always single counterparts. In case of earning itself,

    students in a relationship have around 18 percent higher income than their always single

    counter parts if we control for working hours per week, and 19 percent if we do not control it.

    Due to some omitted variables that were not possible to incorporate in this study (discussed

    in section 4), these results are likely to be overstated. All these results are both statistically

    and practically significant. However, due to the evidence of highest premium among students

    who were previously in a relationship that broke up, we are not clear whether the effect of

    being in a relationship on income is a causal effect or an outcome of selectivity.

  • 8/2/2019 My First Empirical Paper

    19/23

    Page 19 of23

    References:

    Blau, Francine and Andrea Beller. 1988. Trends in Earnings Differentials by Gender,

    1971-1981.Industrial and Labor Relations Review 41 (4), 513-529.

    Becker, Gary S. 1981. Treatise on the family. Cambridge, MA: Harvard University Press.

    Bellas, Marcia. 1992. The Effect of Marital Status and Wivess Employment on the Salaries

    of Faculty Men: The (House) Wife Bonus. Gender and Society 6 (December), 609-622.

    Blackburn, McKinley and Sanders Korenman. 1994. The Declining Marital-Status Earnings

    Differential.Journal of Population Economics 7 (July), 247-270.

    Chun, Hyunbae and Injae Lee. 2001. Why Do Married Men Earn More: Productivity or

    Marriage Selection.Economic Inquiry 39 (April), 307-319.

    Cornaglia, F. & Feldman, N.E., 2010. The Marriage Premium Revisited: The Case of

    Professional Baseball, The London School of Economics.Cornwell, Christopher and Peter Rupert. 1995. Marriage and Earnings.Economic Review,

    Federal Reserve Bank of Cleveland, Q (IV), 10-20.

    Daniel, Kermit. The Marriage Premium, In The New Economics of Human Behavior,

    Mariano Tommasi and Kathryn Ierulli (eds.). New York: Cambridge University Press, 1995.

    Hill, Martha. 1979. The Wage Effects of Marital Status and Children. The Journal of

    Human Resources 14(4), 579-594.

    Krashinsky, Harry A. 2004. Do Marital Status and Computer Usage Really Change the

    Wage Structure?Journal of Human Resources 39 (3), 774-791.

    Light, Audrey 2004. Gender Differences in the Marriage and Cohabitation Income

    PremiumDemography41 (May), 263-284

    Neumark, David. 1988. Employers Discriminatory Behavior and the Estimation of Wage

    Discrimination.Journal of Human Resources 23 (3), 279-295.

    Ribar, David C. 2004. What Do Social Scientists Know about the Benefits of Marriage? A

    Review of Quantitative Methodologies. IZA DP No. 998.

    Rogers, William M. III and Leslie S. Stratton. 2005. The Male Marital Wage Differential:Race, Training, and Fixed Effects. IZA DP No. 1747.

    Waite, Linda J., and Maggie Gallagher. 2000 The Case for Marriage: Why People Are

    Happier, Healthier, and Better Off Financially. New York: Broadway Books.

  • 8/2/2019 My First Empirical Paper

    20/23

    Page 20 of23

    APPENDIX

    Survey Questionnaire

    This questionnaire does not ask for any identification information. So, please feel free.

    1. Do you have your own source of earning?a) Yes b) No 11. Fathers educational qualification:..

    *** If your answer in No, Please skip

    question 2 to 5 and start answering again

    from question 6***

    12. Mothers educational qualification

    ..

    2. For how many months you have your own

    source of income earnining?

    ..

    13. Number of brothers and sisters

    ..

    3. What is your average monthly income?

    ..

    14. Up to HSC, where have you spent the

    major portion of your life?

    a) Dhaka b) Outside Dhaka

    4. What is/are your source(s) of earning?

    a)Tuition b) Others

    15. SSC GPA (without fourth subject)

    .

    5. How many hours you usually work in a

    week for that earning?

    ..

    16. HSC GPA (without fourth subject)

    ..

    6. What is your relationship status?

    a. Single & never in a relationship

    b. Single but previously in relationship that

    broke up

    c. In a relationship / engaged

    17. Present Field of Study:

    a) Social Sciences and Humanities

    b) Business

    b) Science

    7. (If you are presently in a relationship or

    engaged) what is your partners/ girlfriendsincome?

    ..

    18. Currently studying in which year?

    ....................................................

    8. What is your family income per month

    (excluding your own income)?

    ..

    19. Current CGPA (In case of masters

    students use Last Honors CGPA)

    ..

    9. Do you live with your own family?

    a) Yes b) No

    20. Do you receive any regular amount from

    your family?

    a) Yes b) No.

    10. What is rent of the house where your

    family live? (Please write 0 in case of own

    house)..

    21. (If yes) what is the average amount that

    you received from family per month?

    ..

    23. (If yes) what is the average amount that

    you send to your family per month?

    ..

    Thank You

  • 8/2/2019 My First Empirical Paper

    21/23

    Page 21 of23

    Table 1: Definition of the Variables

    Variable Description

    ownsource = 1 if student has own source of earning, = 0 if otherwise

    exper = for how long the student has own source of earning, in months

    inc = students average self-earned income (Tk.) per monthlinc = log(inc)

    weekly_hrs = hours of work per week for that earning

    weekly_hrs_sq = (weekly_hrs)^2

    always_single = 1 if single and never in a relationship,

    = 0 if otherwise

    breakup = 1 if presently single but previously in a relationship that broke

    up, = 0 if otherwise

    relationship = 1 if presently in a relationship,

    = 0 if otherwisespouseinc = spouses income (Tk.), if presently in a relationship

    familyinc = family income excluding students own income (Tk.)

    fathereduc = fathers education in years

    mothereduc = mothers education in years

    siblings = number of siblings

    dhaka = 1 if up to HSC major portion of life spent in Dhaka,

    = 0 if otherwise

    sscgpa = SSC grade point average (in 5.0 scale)

    hscgpa = HSC grade point average (in 5.0 scale)business = 1 if presently studying business,

    = 0 if otherwise

    science = 1 if presently studying science,

    = 0 if otherwise

    scosci_humnts = 1 if presently studying social science or humanities,

    = 0 if otherwise

    year = years spent in university

    cgpa = university cumulative grade point average (in 4.0 scale)

    allowance = amount of allowance received (Thousand Tk. per month)support = amount of money sent to family (Thousand Tk. per month)

    netallowance = allowance support(Thousand Tk. per month)

  • 8/2/2019 My First Empirical Paper

    22/23

    Page 22 of23

    Table 2: Summary Statistics

    Variable Obs Mean Std. Dev. Min Max

    ownsource 285 0.554 0.498 0 1

    exper 275 13.87 18.75 0 84

    inc 283 3364.31 4622.31 0 35000

    weekly_hrs 276 8.00 11.87 0 100

    always_single 278 0.594 0.492 0 1

    breakup 278 0.137 0.344 0 1

    relationship 278 0.270 0.445 0 1

    spouseinc 38 3881.58 5729.76 0 25000

    familyinc 263 22742.54 20249.15 0 150000

    fathereduc 273 12.11 4.86 0 21

    mothereduc 269 9.14 4.39 0 18

    siblings 283 3.28 2.05 0 11

    dhaka 284 0.317 0.466 0 1

    sscgpa 279 4.432 0.482 3.13 5

    hscgpa 279 4.394 0.438 2.4 5

    business 283 0.318 0.467 0 1

    science 283 0.233 0.424 0 1

    scosci_humnts 283 0.452 0.497 0 1

    year 282 3.429 1.281 1 5

    cgpa 240 3.287 0.303 2.18 3.9

    allowance 279 2.353 3.679 0 50.000support 281 0.570 2.406 0 31.500

    netallowance 277 1.629 3.629 -31.500 12.000

  • 8/2/2019 My First Empirical Paper

    23/23

    Page 23 of 23

    Table 3: Regression Results

    Standard errorsin parentheses,

    +p < 0.10, *p < 0.05, **p < 0.01

    (d) for discrete change of dummy variable from 0 to 1.

    Dependent Variable:

    (1)ownsource

    (2a)

    linc

    (2b)

    linc

    relationship (d) 0.208**(0.073)

    0.184+(0.104)

    0.192+(0.100)

    breakup (d) 0.138

    (0.087)

    0.316*

    (0.129)

    0.338**

    (0.127)

    fathereduc -0.027**

    (0.010)

    -0.022

    (0.016)

    -0.014

    (0.015)

    mothereduc 0.022*

    (0.011)

    0.030+

    (0.017)

    0.027+

    (0.016)

    siblings 0.005

    (0.016)

    -0.004

    (0.024)

    -0.015

    (0.023)

    lfamilyinc -0.090*(0.043)

    0.145*(0.065)

    0.118+(0.064)

    netallowance -0.050**

    (0.000)

    -0.064**

    (0.000)

    -0.076**

    (0.000)dhaka (d) 0.005

    (0.074)

    -0.065

    (0.111)

    -0.086

    (0.109)

    sscgpa 0.048

    (0.087)

    -0.102

    (0.134)

    -0.170

    (0.130)

    hscgpa 0.029

    (0.098)

    0.360*

    (0.159)

    0.316*

    (0.155)

    cgpa -0.126

    (0.109)

    0.013

    (0.154)

    0.041

    (0.151)

    year -0.009

    (0.030)

    0.032

    (0.049)

    0.049

    (0.047)

    science -0.040(0.084)

    0.076(0.132)

    0.101(0.129)

    business 0.108

    (0.082)

    0.167

    (0.124)

    0.180

    (0.122)

    exper 0.002

    (0.003)

    0.001

    (0.003)

    weekly_hrs 0.007(0.004)

    intercept 1.662**

    (0.612)

    5.515**

    (0.965)

    6.189**

    (0.900)

    Number of Obs. 197 110 114R2 0.325 0.487 0.477

    Adj. R2 0.274 0.399 0.397

    Root MSE 0.416 0.447 0.448