# Who Needs a Friend?How Age and Having Someone one can Count on Explain Subjective Well-Being in India

## What is the final, best-fitting regression model from this study?

### Short answer: A partial proportional odds (PPO) model. I fit the model using the gologit2 user-written program for Stata (Williams, R. 2016).

Where:

 i = an individual survey respondent Yi = SWB outcome category for respondent i j = 1, 2 = (SWB outcome categories – 1) $\alpha$j = the set of constants from all j models Xi = the set of that respondent’s scores on all x1-x11 variables in the model βj = the set of coefficients for that predictor or control variable at the respondent’s SWB level; for variables where the proportional odds assumption was held, that coefficient is independent of outcome level and does not vary by j. Each respondent i has their own independent multinomial distribution

``````//I cleaned the data first and renamed variables
//I am using Stata 16.1 SE.
svyset [pweight=wgt]
//best final ppo/gologit model for 2009
gsvy: gologit2 WB c.logincome i.socialsup c.logage##i.disability i.prosocial i.freedom ///
i.education i.corruptind i.gender i.region i.hindu if year==2009, ///
autofit force
eststo ppo2009
gsvy: gologit2 WB c.logincome i.socialsup c.logage##i.disability i.prosocial i.freedom ///
i.education i.corruptind i.gender i.region i.hindu if year==2019, ///
autofit force
eststo ppo2019
/* References
Gallup (2021). Gallup World Poll [The_Gallup_101521.dta]. Washington, DC, Gallup.
Jann, Ben (2013). COEFPLOT: Stata module to plot regression coefficients and other results. Available from http://ideas.repec.org/c/boc/bocode/s457686.html.
Williams, Richard (2016). GOLOGIT2: Understanding and interpreting generalized ordered logit models. The Journal of Mathematical Sociology, 40:1, 7-20, http://www.tandfonline.com/doi/full/10.1080/0022250X.2015.1112384. */``````

## FAQs

Historically, disability was thought of as being a personal attribute of a human being. For example, one would be considered disabled if one had their left leg amputated. This is consistent with a medical model for understanding disability, and it was the primary conceptualization in most countries until the 1990s, when a more social model began to emerge. The framing of a problem matters because it suggests or limits the potential solutions and interventions. If the problem is a missing left leg and the situation is framed as a tragedy, the logical intervention is charity. If the problem is a missing leg and the situation is framed as a medical issue, the logical intervention is medical treatment. However, if the problem is that someone with such an impairment cannot access a school, for example, the situation is framed as a social issue and the needed intervention is at the level of institutional policy change towards universal design and accessibility. Because after all, in a world full of ramps, a person who uses a wheelchair is not disabled.

More info: In a development context, the United Nations also defines disability socially, as the result of interaction between personal impairments and environmental barriers that hinder one’s participation in society. Reference: The Convention on the Rights of Persons with Disabilities (CRPD), a 2006 human rights treaty signed by 168 nations including India, led by the United Nations Department of Economic and Social Affairs.

I want to share a somewhat oversimplified explanation first. For all variables in the model that are dichotomous indicators, which is all variables except income and age, a positive coefficient means that a score of “yes” on that indicator increases the likelihood of higher categorical levels of the SWB outcome. A negative coefficient means that a score of “yes” increases the likelihood of the present or lower categorical levels of the SWB outcome.

Of course, because this is not a regular ordered logit model, these coefficients sometimes vary depending on the outcome level we’re looking at. Coefficients are the same at all levels only for those terms for which the proportional odds, or the parallel lines assumptions, held. (If that assumption was upheld for all terms, I would have used an ordered logit though, so it’s not that simple to interpret, unfortunately.)

A PPO/gologit model is interpreted by considering the coefficients at each level of the ordered outcome. So in this case, we have three levels of an ordered outcome, from low to high they are suffering, struggling, and thriving, numerically identified by 1, 2, and 3. The regression coefficients reflect the effects of each of the 11 variables in the model, the key predictors and the controls, and the coefficient represents the odds of a respondent being either above the lowest outcome category (of suffering), or of being in the highest outcome category (which is thriving). Revisit the equation for the model too, noting the role of this probability of being “> j”.

This excellent question came up in a live session. Social support was measured using the Gallup World Poll survey item WP27, which asks:

If you were in trouble, do you have relatives or friends you can count on to help you whenever you need them, or not?

It is a binary indicator, and a respondent either has at least one person they can count or, or nobody. The measure does not address magnitude (i.e. how many friends one has). Someone has social support if they respond “yes” to this item.

Share on: