This is the draft of a paper for 2009 Conference of the New Zealand Statistical Association, 3 September 2009. The presentation was a PowerPoint based on it. It is part of a study of the impact of drinking by associates undertaken by the Centre for Social and Health Outcomes, Research and Evaluation (SHORE), Massey University.
Authors: Brian Easton and Ru Quan (Ryan) You (with Sally Casswell and Taisia Huckle)
Keywords: Health; Statistics;
An earlier paper described the 3068 respondent sample survey on which this is based.
This report focuses on drinking behaviour. Respondents were asked about their own drinking and also whether they had associates who were heavy drinkers, as well as their personal characteristics and self-assessments of their personal well being.
The statistical method is to use the various responses to predict the personal well being variables. Insofar as the equation is valid, the impact of each of the independent variables can be assessed. This enables the consequences of drinking or of associating with drink to be evaluated.
The primary purpose of the survey was to assess the effects of drinking associates on aggregate welfare. It also gave some useful insights into the effects of self drinking behaviour.
Each respondent was asked on 12 dimensions of life including satisfaction with life as a whole (sometimes abbreviated to ‘happiness’). The responses were on a 0 to 10 scale. (Casswell, Huckle &You 2009) Note that the personal wellbeing responses are subjective, although there is some evidence that they correspond reasonably well to an objective assessment (where one exists)..
We asked whether the respondent had drunk alcohol in the last year. (We asked also of those non-drinkers were they ex- drinkers. but their responses have not been evaluated yet.) Drinkers were then asked a series of questions which in this report are summarised as
The frequency of drinking occasions (reported here as occasions per week);
The typical number of drinks on each occasion (reported here as 15ml of absolute alcohol drinks).
While there are problems with individual’s recall of their drinking behaviour although because of the way respondents are asked, SHORE surveys elicit far greater aggregate coverage than many other surveys.
Additionally respondents were asked about associates who were heavy drinkers. Their responses were categorised as:
Level 0: no heavy drinkers among associates (71.6%)
Level 1: associated with heavy drinkers not living or occasionally lived in the same household (15.4%)
Level 2: heavy drinkers sometimes living in the same household (7.5%)
Level 3: heavy drinkers living in the same household for more than half of the time (5.5%)
There is also an adjustment for multiple heavy drinking associates.
(For details see Casswell et al 2009)
Additionally they were asked for various (largely objective) personal characteristics. These are described in You & Easton (2009).
The Impact of Self Drinking
The approach is similar to Ryan & Easton (2009) with the following adaptions.
– we need to avoid the situation of a drinker who drinks nothing.
– the amount and occasions variables are cardinal;
– the occasions and amounts variables are related, at least through having a common intercept.
We found the following equation from worked well, where
Y is the normal logit of the dimension of life variable;
Z the other demographic variables;
A is the amount drunk on a typical occasion
F is the frequency of occasions
u is a random variable with the usual properties.
Y = a0 + b*Z + a1A + a2A2 + a3A3 + f1F + f2F2 + f3F3 + u,
and where the as and fs are constants to be estimated – we estimated them differently for men and women.
We found a cubic polynomial satisfactory. In some cases a shorter one would have worked, but for consistency we maintained them all at order three.
Interpreting the data the following should be kept in mind.
– it would be very rare for someone to have been surveyed while they were drinking; this is about responses to drinking when one is not drinking;
– this probably does not measure long term welfare effects of drinking, and certainly does
not measure any long term effects on health (such as cirrhosis of the liver).
We found that as a rule regularity of dinking did not greatly affect personal well being after we controlled for other characteristics) . For instance, consider the impact on the life satisfaction of a woman who imbibes 15ml of absolute alcohol each occasion by the number of days a week she drinks. It rises slightly, suggesting the regular drinker – of one glass – is likely to be happier than the infrequent drinker. Given that the zero is the never drinker, you can see the modest drinker is slightly better off. This low gradient was generally true for men, and for other life dimensions. However it is not always so that drinkers – even moderate drinkers – are better off than never drinkers.
(Note that this assessment does not include any effect of long term physical health nor the impacts of the drinking on others.)
The male response was also largely unaffected by the frequency of drinking, but they reported that they were less satisfied with life than non-drinkers.
A stronger effect comes from looking at the number of drinks per occasion. While moderate drinkers were sometimes higher on the dimension of life measure than non-drinkers, sometimes lower, those that drunk more were always lower than those that drank less except that the functions tended to flatten out for very large drinking, perhaps because we did not have a lot of observations.
When we looked at women assessed by the quantity of their drinking. we found those who frank more were increasingly less satisfied with life. Those who drank a single drink once a week or two drinks on a daily basis were happier than a non-drinker with the same characteristics. Bigger drinkers were less happy.
Men also show as strong a decline although unlike the women all drinkers report a lower satisfaction of life than for never drinkers (on average). This strong downward slope is universal across all dimensions of personal. However, there is not the time/space to present the individual results – in any case this is a statistics conference concerned with general principles of method, rather than detailed results.
Care needs to be taken in interpreting these graphs. It would be easy to misinterpret them as indicating that a person can raise their personal wellbeing by reducing the quantity they drink each occasion, moving back up the graph line.
However seductive as such a conclusion may seem, the graph lines are not causal relationships. They may be thought more of as sophisticated correlations. We cannot tell whether they are saying that the unhappy drink, and the unhappier they are the more they drink, or whether those who drink make themselves unhappy. Probably both effects are then in the curves.
Unfortunately, this is survey data, and it is never easy to identify causal relationships from cross-sectional data.
However, if this research strengthens anyone’s resolve to reduce their daily alcohol consumption to a modest (or even zero quantity) , the decision would be beneficial even if the conclusion cannot be statistically validated.
The Impact of Drinking Associates
Again the procedure is similar to Ryan and Easton (2009) where each person has a set of personal characteristics (vector) Z, with their drinking associates status indicated by (H1, H2, H3) which has Hx either 1 (associates have the characteristic) or 0.
If each’s person level of welfare on some dimension is given by Y as before, and using the same notation. then we estimate assume that
Y = a + b*Z + h1H1 + h2H2 + h3H3 + u,
Once the coefficients have been estimated the impact of heavy drinkers on the personal wellbeing of New Zealanders can be calculated. Here is an example for Satisfaction with Life.
The mean for satisfaction with life was 7.5 out of 10. (logit(.75) = 1.098).
The coefficients in the equation for H1, H2, H3 were – 0.048, -0.111, -0.353, and they apply to 15.4%, 7.5%, 5.5% of the population, giving a weighted average change of -.035.
It follows that were there no heavy drinking associates the mean of the scored satisfaction with life would 1.098 + 0.035 = 1.133 in logit form, or 0.756 (Since Probit(.756) = 1.133), or 0.85 percent higher.
This gives some credibility to Easton’s guestimate that associates reduced the quality of life by about 1 percent (or it shows that he was very lucky). Easton (1995) too that not all the dimensions of life experience a similar reduction and religious and spiritual values actual rise. ‘God our help in ages in past’?
The SHORE Survey has provided a rich data base. While its primary purpose is to evaluate the impact on the associates of heavy drinkers, it can also be used to investigate the impact of self-drinking and demographic characteristics on personal wellbeing. In two of the cases – the impact of heavy drinkers on others and impact of demographic characteristics on personal welfare, the conditions are probably such that we can say with some confidence that we are observing causal relationships. However in the case of the self-drinking the causality is likely to be in both directions and we are left with intriguing speculations we cannot verify.