Desktop version

Home arrow Business & Finance

  • Increase font
  • Decrease font


<<   CONTENTS   >>

Model specification

A binary variable, У, was constructed from the WVS responses to the happiness prompt - very happy, quite happy, not very happy, not at all happy - which took the value 1 for respondents who said they were “very happy” or “quite happy” (hereafter, “happy”) and the value 0 for respondents who said they were “not very happy” or “not at all happy” (hereafter, “unhappy”). Table 2.1 shows that, of the 5,580 respondents in India to this

Table 2.1 Happiness and life satisfaction in India and South Africa by social group

Percentage of Respondents in the Categories

Unhappy

Happy

Low

Satisfaction

Moderate

Satisfaction

High

Satisfaction

India

Forward Castes

10.5

89.5

37.9

30.0

32.2

Other Backward Classes

14.3

85.7

41.2

33.3

25.5

Muslim

28.0

72

65.3

19.9

14.8

Dalits

16.8

83.2

40.3

27.6

32.1

Scheduled

Tribes

20.9

79.1

36.4

27.5

36.1

Total

15.7

84.3

38.2

29.4

32.4

South Africa

White

5.4

94.6

23.4

40.9

35.7

Black

26.6

73.4

51.0

28.9

20.1

Coloured

11.2

88.8

39.2

30.2

30.6

Asian

15.8

84.2

34.8

33.1

32.1

Total

19.5

80.6

42.7

31.9

25.5

Source: Own calculations from WVS Longitudinal.

question, 84.3% said they were happy while of the 11,299 South African respondents, 80.6% felt similarly.

In terms of a breakdown of happiness by social group, 89.5% of FCs, 85.7% of OBCs, 83.2% of Dalits, 79.1% of STs, and 72% of Muslim respondents said they were happy. In terms of racial group, 94.6% of Whites, 88.8% of Coloured persons, 84.2% of Asians, and 73.4% of Blacks said they were happy. So, in India and in South Africa, there was prima facie evidence of a group hierarchy to happiness: in India, persons from the FCs were at the top and Muslims were at the bottom of the happiness scale, while in South Africa, Whites were most, and Blacks were least, likely to say they were happy.

A ternary variable, Z, was constructed from the 10-point WVS responses to the life satisfaction prompt. These responses were split into three quantiles such that Z took the values: 1 if the WVS responses were in the lowest quantile, the value 2 if the WVS responses were in the next quantile, and the value 3 if the WVS responses were in the highest quantile. In terms of life satisfaction, the values of Z are, hereafter, taken as representing “low satisfaction” for Z = 1, “moderate satisfaction” for Z = 2, and “high satisfaction” for Z = 3.

Table 2.1 shows that in India and South Africa, respectively, 38.2% and 42.7% of respondents expressed low satisfaction while, respectively, 32.4% and 25.5% expressed high satisfaction. In terms of social groups, the striking feature for India was that nearly two out of three Muslims expressed low life satisfaction with 28% of them expressing unhappiness. For South Africa, the striking feature was that nearly one in two Black persons expressed low life satisfaction, with 27% of them expressing unhappiness.

Conclusions about the link between persons’ social group, on the one hand, and their happiness/life satisfaction on the other, based on the raw data presented in Table 2.1, could misstate the relationship because they ignore the effect of other, non-social group factors which could also have affected feelings of happiness/satisfaction. For example, two persons belonging to the same social group may have different levels of education or income or be of different ages, and these differences could influence whether they were happy. If that were so, then some of the observed strength of the social group-happiness relation might be due to the fact that persons in some groups were, on average, better educated/younger/richer than persons from other groups. A relation between social group and happiness/satisfaction could only be substantiated if such a relation could be shown to exist after controlling for non-group factors. For example, Dalits and Muslims could be less happy/satisfied than FC persons not for reasons of caste or religion per se but - if income and education affected happiness/satisfaction positively - because Dalits and Muslims were poorer or had less education than those from the FCs.

For an estimation sample comprising N persons (indexed, i= 1,... ,N), the happiness equation was estimated using logit methods since the dependent variable, У, took binary values: Y, = 1 if respondent i was “happy”; Y, = 0 if respondent i was “unhappy”. The life satisfaction equation was estimated using multinomial logit methods since the dependent variable, Z, took three values: Z, = 1 if respondent i’s satisfaction was low, Z, = 2 if it was moderate, and Z, = 3 if it was high.

Logit and multinomial logit models

Under a logit model:

where X; = = 1,..., K} represents the vector of observations, for

person/, on К happiness influencing variables and $ = = 1,..., Kj

is the associated vector of coefficient estimates.

In a multinomial logit model with / (in this case, J=3) mutually exclusive possible outcomes, indexed, j = 1,...,/, for each individual i, indexed i = 1, . . . , N, the dependent variable Z, is defined as taking the value j for individual i (i.e., Z, = /) if outcome j occurs for individual /'.

If outcome / is taken as the base outcome, the multinomial logit represents, for each individual (/ = 1, . . ., N), the logarithm of the odds ratio of outcome j {j = 1, 1) - to the base outcome, J - as a

linear function of К determining variables (indexed, k = 1 ... K), with Xjk representing the value of variable k for individual i:

N

where = Pr(Z, = /'), ^ pit = 1 and (3jk are the coefficients associ- z=i

ated with ;th outcome for the &th determining variable, with by definition, f3jk = 0 (k = 1,..., K). The assumption is that these coefficients do not vary across the individuals in the sample.

Following the advice contained in Long and Freese (2014), the results from the estimated happiness equation (equation (2.1)) and the estimated life satisfaction equations (equation (2.2)) are presented in the form of predicted probabilities (i.e., Pr( Y( = 1) for the happiness equation and Pr(Z, = j), j = 1, 2, 3 for the life satisfaction equations) computed from the estimated coefficients. This is made possible by using a suite of options associated with the powerful margin command, available in STATA vl4.0 onwards.6 This is because the logit and multinomial estimates do not have a natural interpretation per se, and they exist as a basis for computing more meaningful statistics which are the predicted probabilities Pr(Y) = 1) and Pr(Z, = 1), Pr(Z, = 2), Pr(Z, = 3).7

 
<<   CONTENTS   >>

Related topics