culture and survey measurement - uic | cuppa
TRANSCRIPT
1
Culture and
Survey Measurement
Timothy Johnson
Survey Research Laboratory
University of Illinois at Chicago
2
What is Culture?
the “shared elements that provide the standards for perceiving, believing, evaluating, communicating, and acting among those who share a language, a historic period, and a geographic location” (Triandis, American Psychologist, 1996).
3
Unpackaging Culture
Research has focused on racial, ethnic, or cross-national differences
With few exceptions, mechanisms responsible for these identity group differences have been unexplored
4
Some Models of Culture
Hofstede (2001)
Schwartz (1992)
Triandis (1996)
Ingelhart (1997)
Trompennars and Hampen-Turner (1998)
5
Geert Hofstede:
Culture’s Consequences
Individualism-Collectivism
Power Distance
Uncertainty Avoidance
Masculinity and Femininity
Long- vs. Short-Term Orientations
6
Individualism vs. Collectivism
Self identity and personal goals vs. norms, obligations and duties
In-groups vs. out-groups
Individualists more commonly make cost-benefit decisions
Collectivists more likely to monitor other’s behaviors and feelings
7 8
Individualist
9
Individualist Collectivist
10
Shalom Schwartz:
Cultural Value Orientations
Identifies 3 bipolar dimensions of culture; each represents an alternative resolution to problems that confront all societies:
Embeddedness vs. autonomy
Hierarchy vs. egalitarianism
Mastery vs. harmony
11 12
Harry Triandis:
Cultural Tightness
Reflects emphasis on heterogeneity and surveillance
13
Horizontal vs. Vertical
Social Structures
Vertical cultures
Emphasize social hierarchies
Horizontal cultures
Emphasize egalitarianism
14
Other Dimensions of Culture
Analytic-linear vs. holistic-dialectical cognitive styles
Communication processesContext requirements
Nonverbal behavior
Self-disclosure
Social participationHistorical experience
Social distance
15
1. Reliability
2. Validity
3. Equivalence
Elements of Social Measurement
in Cross-Cultural Research
16
1. Calibration
2. Category
3. Complete
4. Conceptual
5. Construct
6. Construct operationalization
7. Content
8. Contextual
9. Credible
10. Criterion
11. Cross-cultural
12. Cross-language
13. Cross-level
14. Cross-national
15. Cultural
16. Data
17. Definitional
18. Direct
19. Ethnographic
20. Exact
21. Experiential
22. External
23. Factor
24. Factorial
25. False
26. Formal
27. Foreign language
28. Full
29. Full measurement
30. Functional
31. Grammatical-syntactical
32. Group
33. Idiomatic
34. Interpretive
35. Indicator
36. Instrument
37. Item
38. Language
39. Lexical
40. Linguistic
41. Literal
42. Logical
43. Meaning
44. Measure
45. Measure equivalence
46. Measurement equivalence
47. Measurement unit
48. Metaphorical
49. Metric
50. Motivational
51. Normative
52. Operational
53. Operationalization
54. Partial
55. Procedural
56. Pseudo
57. Psychological
58. Psychometric
59. Relational
60. Relative
61. Response
61. Response
62. Rough
63. Sampling
64. Scale
65. Scalar
66. Scalar/metric
67. Situational
68. Stimulus
69. Structural
70. Substantive
71. Syntactic
72. Technical
73. Text
74. Theoretical
75. Translation
76. Translational/ Linguistic
77. True-score
78. Verbal
79. Vignette
80. Vocabulary
Types of Equivalence
17
Question Comprehension/
Interpretation
Emic (culture specific)
Etic (pancultural)
18
Category Fallacy
Assuming a question or concept is universally understood when in fact understanding is culturally conditioned
19
The meaning of words may
evolve faster than most other
aspects of culture.
Triandis (2004) chapter in Comparing Cultures, Dimensions of Culture in a Comparative Perspective
20
“In this question, what does the
word ‘stress’ mean to you?”
0
5
10
15
20
25
30
African
American
Mexican
American
Puerto
Rican
White
Health Problems
Social Problems
21
Probability of Comprehension Difficulty by
Race/Ethnicity
0.066
0.086*0.099*** 0.093**
0.00
0.02
0.04
0.06
0.08
0.10
0.12
White African American Mexican
American
Puerto Rican
Race/Ethnicity
Pro
ba
bil
ity
of
Co
mp
reh
en
sio
n D
iffi
cu
lty
* p<.05; **p<.01; ***p<.001 significantly different from White
22
Memory Retrieval
Episodic vs. semantic search strategies
Memory cues
23
Ji, Schwarz & Nisbett (2000)
Collectivist societies attend more closely to the behavior of others, resulting in memories for behaviors of others that Americans can only estimate
24
Accessibility
Anchoring
Response formatting
Response styles
Judgment Formation/Mapping
25
Comprehension Difficulty by Response Format and
Ethnicity
0.00
0.05
0.10
0.15
0.20
0.25
White African
American
Mexican
American
Puerto Rican
Race/Ethnicity
Pro
bab
ilit
y o
f
Co
mp
reh
en
sio
n
Dif
ficu
ltie
s
Yes-No Vague Quantifier Numeric
26
Extreme response styles
Acquiescence
Non-differentiation
Measurement Artifacts in
Survey Research
27
Self presentation
Social desirability
Interviewer effects
Response Editing
28
% Uncomfortable Discussing Alcohol Use
with Interviewers from Same/Different
Cultural Groups
0
5
10
15
20
25
30
35
40
African
American
Mexican
American
Puerto
Rican
White
Same Culture
Different Culture
29
A. Question Development Phase
B. Questionnaire Pretesting Phase
C. Data Collection Phase
D. Data Analysis Phase
Available Methods for Addressing
Cross-Cultural Equivalence
30
Question Development Stage
1. Expert consultation/collaboration
2. Ethnographic and other qualitative approaches
3. “Good” question-wording practices
4. “Good” translation practices
31
Good Question Wording Practices
Avoid vague quantifiers
Avoid items with ambiguous or dual meanings
Avoid hypothetical questions
Use simple terms that are similarly understood
Use clear and precise time references
Avoid questions with highly abstract concepts
See also recommendations by Brislin (1986)
32
Other Strategies
Use multiple indicators
Use both etic and emic items
Use KISS method: dichotomous response options
Likely results in loss of precision
Substitute numerical or other nonverbal scales, although:
Many numeric scales are also difficult to use
Numeric scales also not invariant in meaning across groups
Beware of lucky and unlucky numbers
Generally do not completely eliminate word use (e.g., semantic differentials still label end-points)
Alternative numbering schemes can influence reporting
Translation: When things go wrong…
• Pepsi's "Come Alive With the Pepsi Generation" translated into "Pepsi Brings Your Ancestors Back From the Grave" in Chinese.
• Frank Perdue's chicken slogan, "It takes a strong man to make a tender chicken" was translated into Spanish as "it takes an aroused man to make a chicken affectionate."
33
Traditional Approach:
Simple Direct Translation
• Simple, unidirectional translation of survey instrument from a source language into one or more target languages.
• a.k.a., one-way translation
• Now considered an unacceptable approach• Not considered informative because it does not produce objective information about translation quality or specific problems
• A commonly referenced alternative is the Back-Translation procedure
Back-Translation
• a.k.a., double translation
• Described by Brislin (1970) – heavily cited
• Basic procedure calls for a bilingual person to translate a source questionnaire into a target language
• A second bilingual person translates the translated version back into the source language without knowledge of the original instrument.
• The initial and revised versions of the source language version are then compared, discrepancies are identified, and final revisions are made (second round sometimes necessary)
Terms that presented difficulty in translation (Weidmer et al., 1999)
Original EnglishAlternative wording used in final Spanish
version
Back-Translation
Health insurance plan Plan de seguro médico Medical insurance plan
Health provider Profesional de salud Health professional
Rating/rate Calificación/califica Grade/grade
Usually Normalmente Normally
Preventive health steps Medidas de salud preventiva
Preventive health measures
Listen carefully Escucharon atentamente Listen attentively
Health care Altencón médica Medical attention
Prescription medicine Medicamentos recetados Prescribed medicaments
Male or female Niňo o niňa/hombre o mujer
Boy or girl/man or woman
Background Ascendencia Ascendancy
Grade Aňo Year
School Estudios studies36
Alternatives to Back-Translation
• Modified direct translation
• Modified back-translation strategies
• Parallel blind technique
• De-centering strategy
• “Triandis” procedure
• Translation by committee
Current Good Practice for Translation (Harkness, 2008)
• Range of expertise needed: survey design, substantive understanding of content, source/target language competence, knowledge of local fielding situation.
• Team approaches.
• Translation teams should include (1) those who translate, (2) those who review translations, (3) and those who take final decisions on versions (adjudicators).
• Translators should be skilled practitioners who have received training and should normally translate out of the source language into their strongest language.
• Is better to use several translators
• Where possible, each translator should make a draft translation.The alternative is to have each translator do a section.
39
Questionnaire Pretesting Phase
1. Cognitive interviews/structured probes
2. Comparative response scale calibration
3. Comparative behavior coding
4. Compare alternative data collection modes
5. Use of comparative vignettes
40
Behavior Codes Used to Identify
Comprehension Problems
Clarification (unspecified): uncertainty about question, unclear if problem is related to construct or context.
Clarification (construct): request for repeat or clarification of question, or statement indicating uncertainty about question meaning
Clarification (context): uncertainty about question meaning within the context of the question as stated
Clarification (time frame): uncertainty regarding question time frame.
Clarification (rewording): rephrases question before answering.
41
Data Collection Phase
1. Use multiple indicators
2. Use both emic and etic questions
3. Respondent/interviewer matching
42
Data Analysis Phase
Common Goals of Cross-Cultural Analyses:
• Verify interpretive equivalence of measures
• Conduct substantive analyses using equivalent measures and procedures
43
Classification of Analysis Strategies by
Sophistication of Statistical Technique
Elementary and easy-to-use techniques
Less elementary but still easy-to-use techniques
Advanced techniques requiring some expertise and effort
44
Elementary and Easy-to-Use Techniques
Comparison of means (or some other methods of central tendency) of different items across countries
Comparison of correlations between the items under investigation with measures of the underlying dimension or with a benchmark items that can be assumed to represent the dimension
Comparison of correlations between the items under investigation with external variables which are assumed to influence them (or are influenced by them)
45
Less Elementary But Still Easy-to-Use
Techniques
Traditional Item Analysis
Exploratory Factor Analysis
Correspondence Analysis
Multidimensional Scaling
Applying Statistical Controls
46
Advanced Techniques Requiring Some
Expertise and Effort
Multilevel (Hierarchical Linear) Modeling
Multiple group confirmatory factor analysis
Structural equation models
Item-Response models
Multi-Trait Multi-Method analyses
47
Multiple Group Confirmatory Factor
Analysis (CFA)
Can be used to compare factor structure of sets of survey questions across multiple population groups
Several advantages over exploratory factor analysis:
CFA enables overall assessment of model fit, and assessment of the cross-group equivalence of individual items.
48
Structural Equation Models
While confirmatory factor analysis focuses on the measurement model, structural equation modeling considers also the structural model
Can evaluate similarities and differences in theoretical models across multiple cultural groups
49
Understanding Probe
ComprehensionDifficulties
Answering
Probe
Memories
Probe
Certainty
Probe
Question Threat Probe
Embarrassment
Probe
MemoryDifficulties
Social
DesirabilityConcerns
Understanding Probe
ComprehensionDifficulties
Answering
Probe
Memories
Probe
Certainty
Probe
Question Threat Probe
Embarrassment
Probe
MemoryDifficulties
Social
DesirabilityConcerns
Understanding Probe
ComprehensionDifficulties
Answering
Probe
Memories
Probe
Certainty
Probe
Question Threat Probe
Embarrassment
Probe
MemoryDifficulties
Social
DesirabilityConcerns
WHITE AFRICAN AMERICAN LATINO
50
Multilevel (Hierarchical Linear)
Modeling
Permits analyses of individuals nested within cultural groups
Effects on dependent variables of measures assessed at multiple levels of analysis (i.e., respondent, culture, question) can be examined
Recent book by van de Vijver et al.
51
HLM Regression Model Estimates of Fixed Effects for Comprehension and Mapping Difficulties
Comprehension Difficulties Mapping Difficulties
Equation 1 Equation 2 Equation 3 Equation 4 Fixed Effect Coeff. SE Coeff. SE Coeff. SE Coeff. SE
Comprehension Difficulty 0.53** 0.09 Mapping Difficulty 0.52** 0.09 Question characteristics Question Length 0.0004 0.02 -0.0003 0.02 0.10** 0.03 0.10** 0.03 Question Length Squared 0.0001 0.0003 0.0001 0.0003 -0.002** 0.001 -0.002** 0.001 Reading Level 0.45** 0.06 0.43** 0.06 0.18** 0.06 0.16** 0.06 Reading Level Squared -0.02** 0.004 -0.02** 0.004 -0.01 0.004 -0.005 0.004 Abstraction Level Ref=(Concrete)
Most Abstract 0.71** 0.12 0.71** 0.12 0.03 0.15 -0.01 0.14 Somewhat Abstract 0.11 0.08 0.11 0.08 -0.01 0.08 -0.02 0.08 Response Format Ref=(Numerical Open-ended)
Yes/No -1.26** 0.10 -1.11** 0.10 -3.89** 0.22 -3.83** 0.22 Verbal labels -0.99** 0.10 -0.87** 0.10 -1.94** 0.12 -1.89** 0.13 Qualified Judgment 0.28** 0.08 0.27** 0.08 0.20* 0.09 0.19* 0.09 Respondent Characteristics Age 0.01+ 0.01 0.01 0.01 0.02** 0.004 0.02** 0.004 Education -0.05 0.04 -0.05 0.04 -0.03 0.03 -0.03 0.03 Race/Ethnicity (Ref=White) African American 0.31** 0.12 0.31** 0.12 0.01 0.11 0.001 0.11 Mexican American 0.50** 0.12 0.50** 0.12 0.07 0.12 0.04 0.11 Puerto Rican American 0.40** 0.14 0.41** 0.14 0.05 0.11 0.03 0.11 Gender (Male) 0.05 0.09 0.05 0.09 0.13 0.08 0.12 0.08 Intercept -4.28** 0.35 -4.30** 0.35 -3.72** 0.31 -3.69** 0.31
**p<.01 *p<.05 +p<.10 52
Preliminary Conclusions
1. Culture matters.
2. Should assume variability until proven otherwise.
3. Consider how measurement variability might influence results.
4. There are many tools now available to address this problem prospectively.
5. But, no magic solutions.
Thank You.
53