Measuring the Impact Of Gambling

Paper to the Wellington Statistical Group, 2 February, 2009.
(With Quan (Ryan) You Analyst, Centre for Social and Health Outcomes Research and Evaluation (SHORE), of Massey University.)
 

Keywords: Health; Statistics;
 

In 2007 the Ministry of Health commissioned Massey University’s Centre for Social and Health Outcomes Research and Evaluation (SHORE) and Te Ropu Whariki to provide a survey were which would information on the impacts of gambling. The first author of this paper (Easton) was a consultant in both the design of the study and the interpretation of its results while the second (You) was in charge of the statistical analysis of the entire survey,
 

The study’s report is >Assessment of the Social Impacts of Gambling in New Zealand , primarily written by En-Yi (Judy) Lin and Sally Casswell, available on SHORE’s website. This paper focuses on the section on the social costs of gambling, on which the authors worked together.
 

The Survey
 

The survey collected quantitative measures which assessed the negative and positive impacts of gambling experienced by the gambler and by significant others (such as family and friends). The survey collected data on the impacts of gambling from three different ethnic groups within New Zealand: Maori, Pacific peoples and Chinese/Korean peoples.
 

The total sample size of the survey was 7010 and the survey consisted of
            1) a general population sample of 4650 respondents, and
            2) over-samples to allow for separate analysis based on 1000 respondents each for the Maori, Pacific and Chinese and Korean samples.
 

Data collection took place from May 2007 to November 2007 using the SHORE and Whariki in-house Computer Assisted Telephone Interview system (CATI). The response rate was 62% for the general population sample, 74% for the Maori sample, 64% for the Pacific sample and 62% for the Chinese/Korean sample.
 

The survey found that participation, in the previous 12 months, in gambling (excluding raffles) in the general population was 61.8% (60.6% – 62.9%). More than half the population had engaged with Lottery products while participation in other modes of gambling was much lower, with fewer than 10% betting at a racetrack or at the TAB. Electronic gaming machines (EGMs – a.k.a. ‘pokies’) were used by 4% in clubs, 8% in bars/pubs and 8% in the casino (with some overlap). Newer gambling opportunities (such as text messaging and the internet) were used by less than 1% of the sample. There is much more that the survey found. For instance different ethnic groups have different patterns of gambling modes.
 

As well as collecting material on people’s gambling habits and how it affects their close associates, and on all the social characteristics a telephone interview could reasonably ask, the survey also asked the respondents to rate themselves on a five category scale according to thirteen domains of life.
 

For instance, in the case of the state of satisfaction with life, survey respondents were asked
Taking everything into account, how satisfied or dissatisfied are you with life in general these days?
            – very satisfied
            – satisfied
            – neither satisfied or dissatisfied
            – dissatisfied
            – very dissatisfied.
 

The responses were dominated by the first two categories with 39% saying they were very satisfied and 53 percent that they were satisfied. Similar questions were asked for the other domains. (Note that the responses are ordinal; that the statistical analysis did not impose a cardinal metric on the responses.)
 

Not surprisingly, there were associations between respondents self-assessments on these domains, Few of the conclusions are surprising. (Table 1)
 

On four of the thirteen dimensions, higher levels of gambling are associated with lower levels on the life dimension, but on the other hand for nine they are not. On eight of the dimensions higher gambling outlays are associated with lower levels but, perhaps surprisingly, financial situation is one of the five for which this does not apply.
 

The gambling modes are more mixed. We can think of explanations of why some gambling modes are associated with lower levels of welfare, others with higher levels but it is worth noting that lottery products do not appear anywhere. To presage a later finding, Electronic Gambling Machines in bars – ‘pokies’ – appear in the final column – nine times, and always negatively.
 

Of course correlation is not causation, and indeed in some cases the causal direction is unclear. (For instance, it is possible that those who are better off go to race tracks.) Surveys are not good means of assessing causation, as we expand below. But we certainly know more about the patterns and patrons of gambling as a result of this survey.
 

Table 1: Impacts of gambling on domains of life (statistically significant at 5% level)

<>Domain of Life
<>Gambling levels
<>Gambling losses
relative to income
<>Gambling modes
<>  <>Higher levels =
<>Higher loss to income ratio
<>  <>Physical health
<>poorer physical health
<>poorer physical health
<>poorer physical health
Casino tables
EGMs in bar/casino
better physical health
Race track


<>Mental well being
<>poorer mental well being
<>poorer mental well being
<>poorer mental well being
EGMs in bar/ casino
Casino tables
TAB
<>Relationships with family/friends
<>No impact
<>poorer relationships
<>poorer relationships
EGMs in bar/ casino
<>Feelings about self
<>poorer feelings about self
<> poorer feelings about self
<>poorer feelings
EGMs in bar/ casino
better feelings
Race track
<>Overall quality of life
<>No impact
<>poorer quality of life
<>poorer quality of life
EGMs in bar/ casino
Casino tables
better quality of life
Poker/card games
<>Overall satisfaction with life
<>lower satisfaction with life
  <>lower satisfaction with life
<>lower satisfaction
EGMs in bar/ casino
higher satisfaction
Race track
<>Financial Situation
<>No impact
<>No impact
<>better financial situation
TAB
Race track
<>Housing situation
<>No impact
<>No impact
<>better housing situation
EGMs in casino
<>Material standard of living
<>No impact
<>poorer standard of living
<>better standard of living
EGMs in casino
Race track
Housie
<>Work performance
<>No impact
  <>No impact
<>poorer work performance
Casino tables
<>Study/training performance
<>No impact
<>poorer study-related performances
<>poorer study –related performance
Poker/card games
<>Care giving – children
<>No impact
<>No impact
<>poorer parent/caregiver EGMs in bar
<>Care giving – elderly
<>No impact
<>No impact
<>No impact

* Some EGMs in Casinos also include clubs.
 

 

The Social and Economic Impact of Gambling
 

The client, the Ministry of Health, was particularly interested in the social and economic impact of gambling. One of us, Easton, has done a lot of work in similar evaluations of the impact of alcohol, illicit drugs and tobacco, and had been on the the International Task Force which had led to the WHO publication International Guidelines for Estimating the Costs of Substance Abuse. There were obvious merits in using the same methodological framework for gambling, although the application proved a little different.
 

The basic stance of the economics approach is that people make the best decisions for themselves, although we shall have to modify this statement. Without going through the details of the theory, when decisions are made in the context of the market the costs to others of an individual’s decisions are usually covered by the price and income effects, so that the impact on others is (broadly) internalised in the individual’s decision. This is a very powerful idea, and seems to be (roughly) true for a wide variety of economic transactions,
 

When a new product comes onto the market – say a gambling mode – the approach assumes that individuals assess the value of the activity to themselves including the income (access to resources) they forgo using it,, and take the new activity up or not as their assessment indicates. Thus, supposing that they do take the activity up, they judge themselves better off after having paid for the resource they utilise.
 

This approach is neutral on the ethics of the activity. It is not that economists are personally immoral, but rather that they try not to impose their values on others. If society prohibits some activity for ethical reasons – for instance, purchases of human organs and children for adoption are prohibited – economists tend to accept that as a given (although they may, as citizens, take a public stance). They may even calculate the cost of resources – if any – that the prohibition affects. But ultimately the economic analysis takes a neutral stance towards the morality or otherwise of the activity or prohibition.
 

On many activities the community has strong and divided views – gambling is but one example. Advocates like to strengthen their arguments by appealing to economic analysis, but that does not mean the analysis justifies, or otherwise, the appropriateness of the activity. It merely points out one consequence of the adoption or proscription of the activity.
 

There are some important caveats to this conclusion. First, the resource costs to society may not be taken into account when the individual takes a decision. For instance, when a person smokes or drinks alcohol it would be unusual for them when making the decision to consider the extent to which their actions may lead to increased expenditure by the public health system, and are therefore a burden on the taxpayer.
 

The economic prescription is usually to ‘internalise’ these external costs; for instance, to impose a tax to cover the external costs so the purchaser in effect takes them into account. Observe that the economic analysis is not saying the activity is morally right or wrong; rather there is a market failure so that the costs to the decision-maker are not properly aligning with the costs to society and that society can make better (more efficient) decisions by the costs better reflecting the true costs to society.
 

(On the whole such externalities are not a major concern in gambling. There are special taxes on gambling, but no one has been able to provide a coherent framework on how or why to impose them.)
 

In recent years economists have become more systematically aware of a some traits of human behaviour may modify the analysis I have just set out.. Because they want to investigate them in a morally neutral way, the behaviour is called ‘time inconsistent’ although it might more popularly called ‘addictive’. Time inconsistency seems important for some gambling.
 

An example of time inconsistency is the person who goes into a pub intending to have a couple of drinks, but imbibes somewhat in excess of that, and the following morning regrets the additional drinking. The regret is nothing to do with having learned something new – like there was a breathalyser check outside the pub. The drinker may have exactly the same information before, during and after the drinking episode, and yet both prospectively and retrospectively regret the actual episode, even though the drinking seems to have been a rational decision at the time.
 

Time inconsistency may happen a lot – as in impulse buying – but in each case to such a minor degree it can usually be ignored for public policy purposes, However, there are some cases where the subsequent regret is great, and yet the activity gets repeated. Hence the notion of addiction.
 

Such ‘irrational’ behaviour appears to apply to some of those involved in gambling. Indeed the phenomenon is sufficiently recognised that some addicts have an arrangement with their local gambling institutions that they be not allowed to enter. They recognise they may exhibit time inconsistent behaviour, and in a more rational state take action to prevent it.
 

Observe that this resolution is a private contract, not a public intervention. However sometimes public interventions can be effective too. It has also been shown that raising taxes on, say, time inconsistent drinkers (in which externalities are already covered by taxation) can lead to an improvement in their long run welfare. In effect they thank the fiscal authorities for reducing their over-drinking by making the cost of the last drink higher than they wish to pay. (Ordinary drinkers, who dont show time inconsistent behaviour, may be worse off by a tax raised for this purpose.)
 

So we cannot be sure what the right public policy is towards time inconsistent behaviour; the phenomenon is raised here as a way of explaining the research findings, for it appears that some people are markedly worse off from their personal gambling, which is not we would expect were people to show time consistent behaviour.
 

The Counterfactual
 

At the core of the economists’ approach is the notion of social cost, where an activity is always measured against a ‘counterfactual’ –  an alternative scenario in which the activity occurs differently (or not at all). The benefit of an activity is measured by the additional resources the counterfactual situation uses compared to the actual situation.
 

A crucial part of economic analysis is that where there are voluntary individual decisions the net cost to society is zero, although we discuss some complications below. However, involuntary consequences of other people’s actions may generate a social cost.
 

For instance, a gambler may ignore the impact of her or his actions on her or his associates. The gambler may turn to crime or have a deterioration in her or his mental well-being as a result of her or his gambling. Each of these has consequences on the welfare of others either directly (as in the case of associates), or indirectly insofar as others suffer from the individuals criminal actions, or the public sector has to supply services for criminal prosecution and punishment or mental health treatment. It is also possible that gambling affects work and study performance and even whether the individual takes the option of working or studying.
 

There are of course, benefits from gambling. Usually the benefits from the voluntary actions, go to the individual, but sometimes they do not. For instance, the benefits to society from taking an educational course may exceed the benefits to the individual. We are not, aware, however, of any such benefits from gambling. (The possibility that gambling improves mental well-being and quality of life is taken into account by individuals when they choose to gamble.)
 

In summary, we look whether there are any involuntary costs in the current situation as a result of gambling, measuring them against an alternative situation. Ite focused on two counterfactuals:
 

No Gambling: This counterfactual assumes that all gambling (in every mode) does not occur. The analysis does not make any assumptions as to how this happens (e.g. by law, or by everybody voluntarily giving up). In that sense the counterfactual is artificial – but that is true for most counterfactuals. Its purpose is to give a sense of the general significance of gambling.
 

Because we wanted to separate out the effects on gamblers from the impact on others, we report the outcome in two divisions:
            – the impact on the gamblers themselves;
            – the impact on those who have a close associate who is a heavy gambler.
 

The second counterfactual looked at the total impact – on gamblers and associates together of the removal of a single mode of gambling. We chose the mode which the preliminary analysis suggested was the most problematic.
 

No Electronic Gaming Machines: In this counterfactual it is assumed that there is no gambling on poker machines. The aim of this counterfactual is to assess the contribution of one particular gambling mode to overall social costs. The mode to illustrate the principle is chosen because both the literature and this study suggested that EGMs are the most socially costly of all modes. This may seem to be the most policy realistic scenario, but it crucially assumes that no EGM gambler switches to another mode (say roulette) nor to any other socially costly activity (say heavy drinking) so that any displaced activity is not socially deleterious. (The activity assessed here is all gambling on machines, whether they are in bars, casinos and clubs.)
 

It is not intended that these counterfactual scenarios are necessarily feasible from a policy perspective. Rather, they allow an exploration of the implications of current gambling by contrasting it with an alternative situation.
 

Material Aspects (Tangibles)
 

One of the features of substance abuse is that it reduces effective GDP. For instance, a drunk driver may wreck a car, thereby reducing the community’s stock of goods and requiring resources to repair or replace it. That is a material loss to the economy.
 

When we looked at the literature and evidence on gambling we concluded these material costs were not high. They are not particularly high for tobacco use either, other than the extra resources that are used by the health system to deal with the illnesses tobacco generates. Yet these medical service costs of tobacco use are far more than applies for gambling, although there are no epidemiological fractions – which are at t he heart of such evaluations – for gambling. So we concluded that there was little point in pursuing the material aspects of the counterfactual scenarios. As in the case of the substance abuse, by far the larger costs were the intangibles – the quality of life of those involved.
 
Measuring the Impact of Gambling on Welfare
 

Suppose we had identical twins, one of whom was allowed to gamble and the other was prohibited from gambling. The standard economic prediction based on rational choice would be that the twin with the opportunity to gamble would make a rational choice which would result in the possibility of a higher personal welfare if he or she chose to use the opportunity to gamble, At the very least he or she would be no worse off than the twin who was not allowed to gamble.
 

It is not practical to do such a study, but we might use the SHORE survey to find pairs who are otherwise similar, but where one gambles and the other does not, and evaluate their respective welfare. More efficiently, the available data cna be pooled, as follows:
 

Each person has a set of personal characteristic X, with their gambling status indicated by G which is either 1 or 0. Then their level of welfare is given by W, where
                        W = f(X, G) + ε,
where the error term, ε, covers for the omitted variables and (hopefully) has the properties to allow us to validly estimate the function.
 

If there was no time inconsistency (and ignoring the effect on gamblers’ associates)
                        f(X, 1) > f(X, 0)
for those who gamble.
 

Given the ordinal nature of the domain of life data (which is the independent variable in the equation), estimating the function requires a non-linear statistical procedure which places the probability of a person with the particular character into each category.[1] So logistic regressions were used to assess the impact of gambling. The demographic variables used to isolate the independent impact of gambling were age, gender, ethnicity, marital status, education qualification, occupational status, income (with a log transformation) and prevalence of other heavy gamblers in one’s life.
 

We then predict the probability of outcome for each person in the survey (on average they will equal the actual outcomes). Aggregating across the entire population gives us the population proportions in each category.[2]
 

We can now carry out exactly the same exercise by setting G = 0 for all members of the population. This is equivalent to testing a counterfactual related to G.
 

We carried out the exercise for all the dimensions of life, although some of the estimated functions did not prove particularly statistically robust. The results for the ‘Overall Satisfaction of Life’. are set out in Table 2. We earlier reported the first data column. The second and third columns assume there is no gambling.
 

The second column suggests there would be a drop off in those ‘very satisfied with life’, presumably because they could not participate in a preferred form of recreation, but that there would also be reductions in the numbers in the bottom three categories where apparently if some were not gambling they would feel more satisfied with life. Overall, much the same number of gamblers in lower categories would move up as those in the higher categories  would move down, with the balance towards gamblers being better off if there were no gambling facilities.
 

Table 2: Satisfaction with Life by Reported State

<>STATE OF
SATISFACTION WITH LIFE
<>ACTUAL SITUATION
<>No-Gambling
<>No-EGMs
<>Gamblers
<>Associates
<>Very Satisfied
<>1245000
<>1240000
<>1262000
<>1259000
<>Satisfied
<>1666000
<>1676000
<>1655000
<>1658000
<>Neither Satisfied or Dissatisfied
<>174000
<>172000
<>171000
<>171000
<>Dissatisfied
<>52000
<>51000
<>51000
<>51000
<>Very Dissatisfied
<>22000
<>21000
<>21000
<>21000
<>TOTAL
<>3160000
<>3160000
<>3160000
<>3160000
<>Net Improvement
<>  <>5000
<>22000
<>22000

Note: Numbers may not sum due to rounding.
 

Column 3 shows a larger and more decisive impact on the associates of heavy gamblers. Some 22,000 of them would be in a higher satisfaction category. This is strong evidence that there is an externality, and that many gamblers do not take into account their impact on others.
 

The final column, which combines the two groups, reports on the effects of the narrower counterfactual of only EMGs being proscribed, but other gambling modes continue to be available. This time there is a rise in the number of people (gambl;ers and associates) who were very satisfied, and a fall in those in the bottom three categories. In all about a net 22,000 people would be in a higher category if EGMs were proscribed.
 

That seem, at first, counterintuitive. By increasing choice – in this case making EGMs available – people feel worse off. But observe that it does explain the previous finding. The people who are better off because they have an opportunity to gamble, are offset by the users of EGMs who are worse off.
 

We can see this even more clearly if we look at mental wellbeing. Survey respondents were asked
Now we are interested in finding out about your mental well-being. In general, in the last 12 months would you say your mental well-being has been …
            – very good;
            – good;
            – adequate;
            – poor;
            – very poor.
 

As in the previous case responses were dominated by the first two categories. Some 54 percent said their mental health was ‘very good’, and another 35 percent said it ‘good’.
 

Table 3: Mental Well-being by Reported State

<>STATE OF MENTAL
WELL-BEING
<>ACTUAL SITUATION
<>No Gambling
<> 

No EGMs

<>Gamblers
<>Associates
<>Very Good
<>1722000
<>1767000
<>1,734,000
<>1767000
<>Good
<>1112000
<>1089000
<>1,104,000
<>1085000
<>Adequate
<>273000
<>255000
<>269,000
<>257000
<>Poor
<>46000
<>42000
<>45,000
<>43000
<>Very Poor
<>8200
<>7300
<>7,900
<>7500
<>TOTAL
<>3160000
<>3160000
<>3,160,000
<>3160000
<>Net Improvement
<>  <>74000
<>18,000
<>69500

Note: Numbers may not sum due to rounding.
 

The changes under the two counterfactuals were larger in magnitude (but of similar direction). This time 74,000 gamblers and 18,000 associates would have been a higher mental health category were there opportunities to gamble; of these 69,500 would have been in a higher category had there been no EGMs, This means that a net 22,500 would have been in a better mental state, if other gambling opportunities as well as EGMs closed down.
 

This is further, and stronger, evidence for time inconsistency of decisions among those who gamble. Some seem to gamble even though it makes them mentally worse.
 

It also shows gambling impacts on the mental wellbeing of non-gamblers, but to a slightly lesser extent than on their life satisfaction.
 

Before the paper discusses the social science and policy implications, statisticians deserve some discussion on metrics and causation.
 

Ordinality and Cardinality
 

The independent variables we have been using are ordinal. The figures for the net improvements report only those who change categories. Were there an underlying cardinal scale, it is possible that some people would have an improvement on it, but not sufficient to change categories.
 

We do not know what cardinal scale – if any – the survey respondents had in mind when they answered the domain of life questions. While we are hesitant to impose one on them, we wanted to get a feel of the magnitude of the changes, including those that remained in each category but felt some improvement.
 

The simplest approach would be to impose a simple scale – say 0 to 5 – on the five categories, and assume that the mean scores in each category do not change. The results are in Table 4.
 

Table 4: Increase in Overall Welfare

<> 

STATE OF
<>No Gambling
<>No EGMs

<>Gamblers
<>Associates
<>SATISFACTION WITH LIFE
<>0.01%
<>0.20%
<>0.18%
<>MENTAL WELL-BEING
<>0.54%
<>0.11%
<>0.48%

Note: Cardinal Scale as explained in text.
 

In each case the removal of gambling, whichever the counterfactual, increases the welfare measure, in a manner similar to the numbers who would change in their categories in Tables 2 and 3.
 

More sophisticated metrics could be imposed on the ordinal scale. They are unlikely to lead to deeper insights, especially if the assumptions to construct the scale are allowed for. In the long run what is needed is a survey which enables the domains of life categories to be mapped onto a cardinal scale.
 

It might be thought the numbers are not large. However, an improvement of the life satisfaction of 22,000 people say, or the mental well being of 92,000 people, might be compared the 400 odd road deaths annually for which a considerable public effort is put into reducing; the social costs of tobacco and alcohol are about an order of magnitude more than those for illicit substance abuse. but we still put a lot of effort into dealing with the problem. Others may well judge that a half a percent (or whatever) improvement in overall mental health is s a worthwhile public health gain.
 

Correlation and Causation
 

Correlation does not prove causation; indeed it is difficult for a cross-sectional survey by itself to say much about causation. Yet much of this paper has been written in the language of causation. The justification will be found in the part of the report which surveys the scientific literature on the effects of gambling, and which suggests that the causation paths are valid.
 

Even so, there remains the possibility that there are also a causal paths in the reverse direction. For instance, suppose a subsector of the population had a gene which made them prone to melancholy and to take up gambling. It is not possible for a telephone survey to identify this gene (supposing it exists), and so its existence is an omitted variable and a potential source of statistical bias. However those who wish to argue the case for such reverse causality, need to provide some evidence for the causal path, rather than argue one exists a priori.
 

A more serious limitation on the research findings may be that the counterfactuals assume that the prohibition on gambling modes does not lead to deleterious displacement behaviour. For instance, perhaps if only EMGs were prohibited their current users might turn to other gambling modes, with for them, similar welfare outcomes; if all gambling was prohibited, perhaps some gamblers would turn to substance abuse (or illegal gambling).
 

Implications for Public Policy
 

It is rare for a single piece of research to be decisive for public policy. The population survey which underpins this paper would make no such claim. Rather it builds on what is known, providing insights into the magnitude of gambling in New Zealand, and quantifying differences between different social groups – by ethnicity, by age and by gender. It has the additional strength that it has paid attention to the impact of gambling on the associates of gamblers.
 

The statistical research reported here has made a similar incremental contribution, demonstrating how gambling appears to have an important impact on the welfare of many New Zealanders. Completely consistently with the literature, it shows many New Zealanders are worse off as the result of gambling, including some who are associates of gamblers – mental health seems particularly affected. But also it draws attention to the fact that were there no opportunities to gamble there are those who would have a lower satisfaction with life.
 

Again consistent with the literature, but perhaps surprising in magnitude, the research shows that electronic gambling machines are a particular problem. A high proportion of those with poorer mental health and lower satisfaction of life use EGMs.
 

Despite the findings from the relevant counterfactual, it does not necessarily follow that EGMs (or indeed all modes, or any other mode, of gambling) should be prohibited; the assumption that there will be no displacement to other socially unsatisfactory activities may not hold. What the research here reinforces is that EGMs provide a substantial challenge which public policy may wish to address.
 

If there is any firm public policy conclusion from this study it is, of course, the need for more research. In particular the opportunities that the data base from the SHORE/Whariki survey of 7010 New Zealanders presents are far from fully exploited. Further statistical analysis will shed more light upon gambling behaviour in New Zealand, but there are also the possibilities of a sensitive exploration of the interdependencies of the domains of life. It will be disappointing if the potential in the SHORE/Whariki data base on gambling to understand the welfare (and health) of New Zealanders is not further explored.
 

Go to top
 

References
[1]  In practice the ordinality requires the estimation of the probability of being in the top category, the top two categories, the top three categories and the top four categories, and subtract to get the probability for individual categories.
[2]  The sample elements were weighted in line with their population proportions for adults over 15.
 

Go to top