Subjective Well-Being

Final report for 2nd R Class by Paul Okopny

March, 2013

Abstract

This paper demonstrates the usage of “Cultural Fit” concept to test wherether cultural values and descrepaces could be a deteminants for Subjective Well-Being. The “usual” SWB determinants, such as health, martial status, and education, are also tested on the ESS data. The research suggests that at country level some of the values are weak determinants for SWB, while cultural descrepancy is not.

Literature Review

Concept of Subjective Well-Being

The term “subjective well-being” (SWB), which is used by psychologists as a synonym for happiness (Lu, 2006), was promoted by Ed Diener. Since 1984 “Dr. Happiness” - as Diener is known by colleagues (Wallis, 2005) - published about 200 works, and his original work “Subjective Well-Being(1984)” has been cited over 1 200 times (Eid & Larsen, 2008).

The study of SWB focuses on how people evaluate their lives: both at the moment and during long periods of time (Diener et. al. 2003). These evaluations include emotional reactions to events, moods (affects); judgements about life satisfaction, fulfilment (cognitive judgement); and satisfaction with different life domains, such as marriage and work (domain satisfaction). SWB researchers have shown that these constructs are separable, and moreover, pleasant (e.g. joy, pride, ecstasy) and unpleasant (e.g. envy, guilt, shame) affects are moderately inversely correlated, but clearly separable (Diener et. al. 1999).

Measuring SWB

In order to measure SWB, the Satisfaction With Life Scale (SWLS) was developed (Diener et. al. 1985). SWLS consists of five items or statements, each in scale from one to seven, resulting in a five to 35 scale. Diener suggests a number of factors that influence overall life satisfaction: social relationships; work, school, or performance in an important role; satisfaction with the self, standing for spiritual life, learning, growth, and leisure (Diener, 2006). Although psychologists developed a lot of other tools to assess SWB, based on peer-reports, observations, emotion-sensitive tasks, and cognitive tasks, self-report assessments are considered reliable, thus SWLS is considered valid and reliable (Diener et. al. 1999) and used as a standard measure of SWB (Eid & Larsen, 2008).

SWB Determinants

Over decades researchers have found relationships between SWB and an individual's income (Diener et. al. 1993), health (Dolan et. al. 2008), age (Horley & Lavery, 1995), martial status, partner's education (Stutzer & Frey, 2006), self-esteem (Zhang, 2005), etc.

Chinese researcher Luo Lu claims that “cultural conceptions of happiness are critical aspects of SWB, which has largely been neglected” (Lu, 2005). Using a number of scales to measure individual's values (beliefs): Independent and Interdependent Self Scales (IISS), Primary Control Beliefs Scale (PCBS), Harmony Beliefs Scale (HBS); and Chinese Happiness Inventory (CHI) for SWB, Lu proposed a “cultural fit index”: an absolute value of the difference between the individual mean on each item and the group mean. The research showed that individual-level cultural values are consistently related to SWB.

Sagiv and Schwarz, researchers from The Hebrew University of Jerusalem in Israel, discovered two corellations between two indexes of SWB (general mental health and positive affect) and achievement, stimulation, and self-direction values (positive correlation); and traditional values (negative correlation). Further, no correlation was found between values and the cognitive index of SWB (Sagiv & Schwartz, 2000).

Research Design

I use the “Cultural Fit” concept, estimate the Cultural Fit Index (CFi) and it's effect into SWB. I expect to find a consistant relationship between CFi and SWB, as I suppose that individuals who do not fit into their reference groups will be constrained in communications and social actions. I operationalize CFi as a mean of values of differences between individual and reference group values. I define an individual's reference group as a group within the same country and age group. I use the following age groups: young (18-24 years), young adult (25-34 years), adult (35-44 years), middle age (45-65 years), and old (65+ years).

I expect to find a consistant relationship between subjective health and martial status and Cognitive Judgement even without controlling for age and country.

I expect to find a consistant relationship between Creativity values (e.g. importance of being creative) and Cognitive Judgement.

I expect to find a consistant relationship between CFi and Cognitive judgement.

Hypothesys

  1. Health, martial status, and education level will be consistantly related to SWB

  2. Creativity values will be consistantly related to SWB.

  3. Cultural Fit indices will be consistantly related to SWB: traditional values discrepancy will be related with SWB in former communist countries, whereas individual values discrepancy will be related with SWB in protestant countries.

Proposed model 1:

To test the first hypothesys I propose the following model:

Proposed model 2:

To test the second hypothesys I propose the following model:

Proposed model 3:

To test the third hypothesys I propose the following model:

Data description

I use the European Social Survey (ESS) data, that can be fetched by anyone. I use the data from the last (the fith) round and all available countries.

Dataset contains 50781 observations from 26 countries. Mean age of respondent is 48.3932.

I define two sub-groups of countries: protestant (Denmark, Great Britain, Nitherlands, and Switzerland) and former communist (Russia, Ukrain, and Bulgaria).

Variables

Socio-demographic profile

Cognitive judgement (vary from 0 to 10):

Creativity Values (beliefs) (vary from 1 to 6, 1 - very much like me)

Cultural fit indices

CFi are calculated as a mean of values of differences betwee individual and group values.

Data analysis

I use the confirmatory factor analisys (CFA) to test hypothesys. CFA allows us to measure or estimate the loadings of certain factors into construct we are investigating. The purpose of CFA is to determine whether the real data fits the hypothetised model or not.

Testing first hypothesys

I define the first model.

model.1 <- "Cognitive =~ happy + stflife\nStatus =~ health + partner + edulvla\nCognitive ~~ Status"

I test the model against the whole data.

model.1.fit <- cfa(model.1, data = cdata)
m.1.fit <- inspect(model.1.fit, "fit")

The model converged with following fit indices: CFI = 0.9713, SRMR = 0.0366, and RMSEA = 0.0833, which is not acceptable.

p.model.1.fit <- cfa(model.1, data = mature.p)
p.m.1.fit <- inspect(p.model.1.fit, "fit")

I test this model against the groups of middle age and old people and get the following fit indices: CFI = 0.9876, SRMR = 0.0215, and RMSEA = 0.0559, which is acceptable.

summary(p.model.1.fit, standardized = T)
## lavaan (0.5-11) converged normally after  47 iterations
## 
##                                                   Used       Total
##   Number of observations                         27028       27645
## 
##   Estimator                                         ML
##   Minimum Function Test Statistic              342.294
##   Degrees of freedom                                 4
##   P-value (Chi-square)                           0.000
## 
## Parameter estimates:
## 
##   Information                                 Expected
##   Standard Errors                             Standard
## 
##                    Estimate  Std.err  Z-value  P(>|z|)   Std.lv  Std.all
## Latent variables:
##   Cognitive =~
##     happy             1.000                               1.912    0.871
##     stflife           1.069    0.013   84.909    0.000    2.044    0.834
##   Status =~
##     health            1.000                               0.602    0.631
##     partner           0.242    0.008   32.073    0.000    0.146    0.303
##     edulvla          -0.582    0.021  -27.754    0.000   -0.351   -0.249
## 
## Covariances:
##   Cognitive ~~
##     Status           -0.769    0.013  -57.002    0.000   -0.667   -0.667
## 
## Variances:
##     happy             1.167    0.041                      1.167    0.242
##     stflife           1.824    0.048                      1.824    0.304
##     health            0.549    0.011                      0.549    0.602
##     partner           0.211    0.002                      0.211    0.908
##     edulvla           1.863    0.017                      1.863    0.938
##     Cognitive         3.656    0.056                      1.000    1.000
##     Status            0.363    0.012                      1.000    1.000
semPaths(p.model.1.fit, "std", "std")

plot of chunk unnamed-chunk-10

This gives a limited support to the first hypothesys and suggests it will hold for certain groups of people (45 + years).

Testing the second hypothesys

I define the second model.

model.2 <- "Cognitive =~ happy + stflife\nIndValues =~ ipcrtiv + impdiff\nCognitive ~~ IndValues"

I test the model.

model.2.fit <- cfa(model.2, cdata)
m.2.fit <- inspect(model.2.fit, "fit")

Model converged normally. Fit indices: CFI = 0.9999, SRMR = 0.0017, and RMSEA = 0.0113.

summary(model.2.fit)
## lavaan (0.5-11) converged normally after  50 iterations
## 
##                                                   Used       Total
##   Number of observations                         48704       50781
## 
##   Estimator                                         ML
##   Minimum Function Test Statistic                7.202
##   Degrees of freedom                                 1
##   P-value (Chi-square)                           0.007
## 
## Parameter estimates:
## 
##   Information                                 Expected
##   Standard Errors                             Standard
## 
##                    Estimate  Std.err  Z-value  P(>|z|)
## Latent variables:
##   Cognitive =~
##     happy             1.000
##     stflife           0.978    0.019   50.532    0.000
##   IndValues =~
##     ipcrtiv           1.000
##     impdiff           0.805    0.027   30.115    0.000
## 
## Covariances:
##   Cognitive ~~
##     IndValues        -0.459    0.012  -37.844    0.000
## 
## Variances:
##     happy             0.726    0.070
##     stflife           2.026    0.068
##     ipcrtiv           0.743    0.029
##     impdiff           1.285    0.020
##     Cognitive         3.604    0.075
##     IndValues         0.869    0.030
semPaths(model.2.fit, "std", "std")

plot of chunk unnamed-chunk-14

This gives the hypothese weak support and shows the relationship between cognitive judgement and creativity values, though this relationship should be investigated in further research.

Testing the third hypothesys

I define two models.

model.3.a <- "Cognitive =~ happy + stflife\nStatus =~ health + partner + edulvla\nFit =~ ind.fit + status.fit\nCognitive ~~ Status + Fit"

model.3.b <- "Cognitive =~ happy + stflife\nStatus =~ health + partner + edulvla\nFit =~ trad.fit\nCognitive ~~ Status + Fit"

I test the first model against the data, containing obesrvations from protestant countries

model.3.a.fit <- cfa(model.3.a, cdata[cdata$cntry %in% protestant_cntr, ])
m.3.a.fit <- inspect(model.3.a.fit, "fit")

Fit indices (CFI = 0.9522, SRMR = 0.0395, and RMSEA = 0.068) do not allow to accept this model. I test this model against the data, containing observations from former communist countries.

model.3.a1.fit <- cfa(model.3.a, cdata[cdata$cntry %in% former_communist_cntr, 
    ])
m.3.a1.fit <- inspect(model.3.a1.fit, "fit")

Fit indices: CFI = 0.9838, SRMR = 0.021, and RMSEA = 0.0386.

summary(model.3.a.fit, standardized = T)
## lavaan (0.5-11) converged normally after  73 iterations
## 
##                                                   Used       Total
##   Number of observations                          6995        7333
## 
##   Estimator                                         ML
##   Minimum Function Test Statistic              367.157
##   Degrees of freedom                                11
##   P-value (Chi-square)                           0.000
## 
## Parameter estimates:
## 
##   Information                                 Expected
##   Standard Errors                             Standard
## 
##                    Estimate  Std.err  Z-value  P(>|z|)   Std.lv  Std.all
## Latent variables:
##   Cognitive =~
##     happy             1.000                               1.386    0.850
##     stflife           1.132    0.032   35.188    0.000    1.569    0.867
##   Status =~
##     health            1.000                               0.475    0.541
##     partner           0.274    0.021   13.327    0.000    0.130    0.266
##     edulvla          -0.914    0.063  -14.525    0.000   -0.434   -0.303
##   Fit =~
##     ind.fit           1.000                               0.323    0.588
##     status.fit        0.538    0.106    5.093    0.000    0.174    0.403
## 
## Covariances:
##   Cognitive ~~
##     Status           -0.384    0.017  -22.295    0.000   -0.583   -0.583
##     Fit              -0.014    0.009   -1.541    0.123   -0.031   -0.031
##   Status ~~
##     Fit               0.040    0.005    7.763    0.000    0.262    0.262
## 
## Variances:
##     happy             0.740    0.053                      0.740    0.278
##     stflife           0.815    0.068                      0.815    0.249
##     health            0.544    0.018                      0.544    0.707
##     partner           0.224    0.004                      0.224    0.929
##     edulvla           1.861    0.035                      1.861    0.908
##     ind.fit           0.197    0.021                      0.197    0.654
##     status.fit        0.155    0.006                      0.155    0.837
##     Cognitive         1.922    0.067                      1.000    1.000
##     Status            0.225    0.018                      1.000    1.000
##     Fit               0.104    0.021                      1.000    1.000
semPaths(model.3.a.fit, "std", "std")

plot of chunk unnamed-chunk-19

I test the second model against the data

model.3.b.fit <- cfa(model.3.b, cdata[cdata$cntry %in% protestant_cntr, ])
m.3.b.fit <- inspect(model.3.b.fit, "fit")

Fit indices (CFI = 0.9501, SRMR = 0.043, and RMSEA = 0.0842) do not allow to accept this model. I test this model against the data, containing observations from former communist countries.

model.3.b1.fit <- cfa(model.3.b, cdata[cdata$cntry %in% former_communist_cntr, 
    ])
m.3.b1.fit <- inspect(model.3.b1.fit, "fit")
summary(model.3.b1.fit, standardized = T)
## lavaan (0.5-11) converged normally after  52 iterations
## 
##                                                   Used       Total
##   Number of observations                          6275        6960
## 
##   Estimator                                         ML
##   Minimum Function Test Statistic              124.201
##   Degrees of freedom                                 7
##   P-value (Chi-square)                           0.000
## 
## Parameter estimates:
## 
##   Information                                 Expected
##   Standard Errors                             Standard
## 
##                    Estimate  Std.err  Z-value  P(>|z|)   Std.lv  Std.all
## Latent variables:
##   Cognitive =~
##     happy             1.000                               2.126    0.871
##     stflife           0.907    0.027   33.732    0.000    1.928    0.754
##   Status =~
##     health            1.000                               0.457    0.492
##     partner           0.160    0.018    8.788    0.000    0.073    0.146
##     edulvla          -0.664    0.050  -13.212    0.000   -0.304   -0.232
##   Fit =~
##     trad.fit          1.000                               0.466    1.000
## 
## Covariances:
##   Cognitive ~~
##     Status           -0.805    0.030  -26.635    0.000   -0.828   -0.828
##     Fit              -0.011    0.014   -0.775    0.439   -0.011   -0.011
##   Status ~~
##     Fit               0.008    0.005    1.498    0.134    0.035    0.035
## 
## Variances:
##     happy             1.435    0.126                      1.435    0.241
##     stflife           2.825    0.113                      2.825    0.432
##     health            0.655    0.022                      0.655    0.758
##     partner           0.243    0.004                      0.243    0.979
##     edulvla           1.615    0.030                      1.615    0.946
##     trad.fit          0.000                               0.000    0.000
##     Cognitive         4.522    0.161                      1.000    1.000
##     Status            0.209    0.021                      1.000    1.000
##     Fit               0.217    0.004                      1.000    1.000

The result shows that there is no consistant relashionship between traditional values fitness and cognitive judgement.

This rejects the hypothesys, whereas suggests that creative and status values fitness has a weak relationship with cognitive judgement in the former communist countries.

Conclusion

Results showed that there is a consistant, yet weak relationship between creativity values and cognitive judgement. The results also showed that the value discrepancy could have a relationship with cognitive judgement, though further researches should investigate this relationship.

I assume that cultural descrepancy could be a significant determinant of SWB for individuals in certain social groups, especially closed ones. At the macro level (country) it's power as determinant is weak and can be often neglected in SWB or other researches.

The difference of cultures shows itself not only in individuals and groups values, but also in the attitude to and tolerance for discrepancy.

I suggest using “Cultural Fit” concept in further SWB (and other) researches as well as other fitness indices (e.g. ethnicity fit, religion fit, language fit, etc.). These concepts, obviosly originated from biology, are worth trying in social science.

Literature