TY - BOOK
T1 - Causal inference methods in gambling research
AU - James, Richard J. E.
AU - Kim, Hyungseo
AU - Hitcham, Lucy
AU - Tunney, Richard J.
N1 - Copyright © Richard J. E. James, Hyungseo Kim, Lucy Hitcham, Richard J. Tunney, work licensed CC-BY-NC-ND.
PY - 2023
Y1 - 2023
N2 - The purpose of this project was to review and utilise methods from other disciplines in the social sciences in order to be able make stronger causal claims using cross-sectional gambling data such as gambling prevalence studies. We focused on the question of whether there is a causal relationship between specific gambling products and individual gambling harms, specifically problem gambling. There has been an existing literature that has looked at this issue, but fails to control for selection biases on engagement with specific gambling behaviours. We reviewed and used three approaches: propensity score matching, coarsened exact matching, and sample selection modelling to account for this limitation. Then, we applied them to examine the relationship between 8 types of gambling activity (online betting, offline betting, pools, scratchcards, lotteries, pools, offline bingo, and slot machines/FOBTs) on problem gambling in 9 British gambling prevalence surveys: the Health Survey for England in 2012, 2015, 2016 and 2016; the Scottish Health Survey in 2012, 2015, and 2016, and the British Gambling Prevalence Study in 2007 and 2010. The results showed most gambling activities were, unsurprisingly, associated with an increased risk of addictive behaviour. However, there was a clear gradient of risk. For some gambling activities (i.e., online gambling, slots), the effect sizes were substantially higher than others (i.e., lotteries). The modelling also highlighted covariates that exert strong effects on both the IV and DV of interest at the same time, such as age. The findings of this project thus highlight the importance of controlling for selection mechanisms that influence both engagement and outcomes of interest. There are some activities (e.g., bingo, pools) where these substantially affect relationships between engagement and harms. Nonetheless, there is clear evidence of associations between certain types of gambling product and indicators of harm when these are controlled for.
AB - The purpose of this project was to review and utilise methods from other disciplines in the social sciences in order to be able make stronger causal claims using cross-sectional gambling data such as gambling prevalence studies. We focused on the question of whether there is a causal relationship between specific gambling products and individual gambling harms, specifically problem gambling. There has been an existing literature that has looked at this issue, but fails to control for selection biases on engagement with specific gambling behaviours. We reviewed and used three approaches: propensity score matching, coarsened exact matching, and sample selection modelling to account for this limitation. Then, we applied them to examine the relationship between 8 types of gambling activity (online betting, offline betting, pools, scratchcards, lotteries, pools, offline bingo, and slot machines/FOBTs) on problem gambling in 9 British gambling prevalence surveys: the Health Survey for England in 2012, 2015, 2016 and 2016; the Scottish Health Survey in 2012, 2015, and 2016, and the British Gambling Prevalence Study in 2007 and 2010. The results showed most gambling activities were, unsurprisingly, associated with an increased risk of addictive behaviour. However, there was a clear gradient of risk. For some gambling activities (i.e., online gambling, slots), the effect sizes were substantially higher than others (i.e., lotteries). The modelling also highlighted covariates that exert strong effects on both the IV and DV of interest at the same time, such as age. The findings of this project thus highlight the importance of controlling for selection mechanisms that influence both engagement and outcomes of interest. There are some activities (e.g., bingo, pools) where these substantially affect relationships between engagement and harms. Nonetheless, there is clear evidence of associations between certain types of gambling product and indicators of harm when these are controlled for.
UR - https://www.greo.ca/Modules/EvidenceCentre/Details/causal-inference-methods-in-gambling-research
U2 - 10.33684/2024.004
DO - 10.33684/2024.004
M3 - Commissioned report
BT - Causal inference methods in gambling research
ER -