RISK CURVES: GAMBLING WITH DATA

Currie and colleagues have conducted a sophisticated series of analyses exploring risk curves for gambling [1]. The use of multiple definitions of gambling problems as well as several measures of gambling intensity allowed for increased confidence in the results found. There are, however, several concerns with the data that merit discussion. These concerns have to do with the measure chosen for frequency of gambling and the implications of missing data on the validity of the risk curves observed. The intent of this discussion is not to undermine this quality research, but rather to allow for further discussion of the implications of the findings in this study. A strength of the study is that several measures were chosen as markers of amount of gambling activity (amount of money spent, percent of gross income and frequency of gambling). The authors chose the defensible position that frequency of gambling could best be assessed by choosing the most frequent type of gambling behaviour and using this measure as a proxy for the frequency of all the person’s gambling behaviour. The example used in the Currie et al. study [1] was that a person who plays the lottery once a week and electronic slot machines every day would be counted as a daily gambler. The difficulty with this choice of measure becomes more easily apparent if the example is used of someone who plays the lottery once a month, plays slot machines three times a month and bets on horses four times a year. What is the frequency of this person’s gambling? Using the most frequent type of gambling behaviour as a definition, the person would be counted as someone who gambles two to three times per month. However, the person clearly gambles more often than this. An alternative measure would be to sum the frequency of each gambling behaviour to generate a composite frequency of gambling measure. However, it should be noted that this alternative measure also cannot be taken as an absolute measure of the number of days a person gambled, because respondents may have engaged in several gambling activities on one day and in no gambling on another (and there is no way to adjust for this in the composite measure). Thus, the data is limited in that there is no way to approach a true measure of frequency of gambling. Fortunately, there are two reasons why this difficulty of definition should not invalidate the low-risk gambling guidelines suggested based on these analyses. First, as the authors note, under-reporting (or in this case, under-estimating) the frequency of gambling merely leads to conservative low-risk gambling guidelines, something which is defensible from a desire to minimize harm. Secondly, the risk curves are themselves robust. The shape of the curves remains similar to those reported in this paper when a composite measure of gambling frequency is chosen and the same data set is employed [2]. More troubling to the validity of the findings is the preponderance of missing data. While the 2002 Canadian Community Health Survey (CCHS) is large (36 984 respondents), only about 11 000 respondents were actually asked the gambling consequence items that formed the basis of the risk–curve analysis. Many of the respondents in the CCHS were not eligible because they did not engage in any one gambling activity more than six times in the last year. This in itself is marginally problematic, as the Currie analysis made the assumption that these low frequency gambling respondents experienced no consequences. However, more problematic is the fact that almost 9000 respondents excluded themselves by stating that they were not gamblers (including respondents who were frequent gamblers). In addition to reducing the sample size available for analyses, this self-exclusion also potentially limits the reliability of the results. This is because a proportion of gamblers (whether frequent or infrequent) who may have experienced each of the gambling consequences were never asked the questions about experience of these consequences. Thus, there is no way of knowing whether the dose–response curves presented are accurate representations of the risks of gambling in the general population because a true, representative population sample is not available for the analyses. These limitations also call into question the validity of the low-risk guidelines that are generated from them. Currie and colleagues are already appropriately cautious in stating the low confidence level of their findings and stress the need for replication before anything more than tentative conclusions can be made. The implications of missing data merely serve to underline the importance of these recommendations for systematic replication of this research.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.