Household gambling expenditures and the Irish recession

Abstract

This article examines the determinants of household gambling expenditures in Ireland and the effects of the recession on these expenditures using a large micro data-set, the Irish Household Budget Survey (HBS). Two gambling expenditures are examined, bookmaker tote betting and spending on the national lottery. Households with an older and a less educated head of household participate in and spend more on both forms of gambling while the presence of children in the households tends to reduce participation and spending in gambling. There is also evidence to suggest that households with an unemployed head of household have a higher likelihood of participation in gambling. The recession has affected the two forms of gambling in different ways. Lottery expenditures appear resilient to the effects of the recession. This is demonstrated in the estimated expenditure elasticities in particular. Bookmaker/tote expenditures have changed from a necessity to a luxury good, while lottery expenditures have increased in their necessity status. This can be explained by the fact that playing the lottery requires less time, knowledge and risk and has potentially greater benefits attached to it in comparison to bookmaker/tote betting. During a recession these factors become much more prevalent.

Keywords:

• Household gambling expenditures
• log normal model
• income elasticities
• economic recession

Constraints on publishing

No constraints on publishing were reported by the author.

Notes on contributor

John Eakins currently works as a lecturer in the School of Economics at University College Cork, Ireland. His research interests mainly focus on the economics of household decision-making and he has published studies in this area on gambling, sports participation, energy use and transport demand.

Notes

1. A small number of other gambling categories exist in the HBS including football pool stakes and bingo stakes but the expenditures on these categories are very small. Casinos are illegal in Ireland under the Gaming and Lotteries Act, 1956, although many private clubs offer casino-like facilities which exist through loopholes in the law. These private clubs are generally small establishments.

2. The CSO stated that information on income in the HBS ‘is used primarily for categorical purposes (e.g. for analysis of households according to different levels of disposable income) rather than the provision of information on income levels’ (Central Statistical Office, 2013, 41).

3. The Tobit model has been applied previously to the analysis of gambling expenditures by Livernois (1986), Clotfelter and Cook (1987), Scott and Garen (1994), Farrell and Walker (1999), Worthington (2001), Worthington et al. (2007) and Pérez and Humphreys (2011).

4. Other hurdle models that are commonly applied include the truncated normal model where the second stage is modelled using a truncated normal regression model. Another version, commonly referred to as the double hurdle model, exists where the simultaneous estimation of the likelihood function from the first and second stage is carried out; see Humphreys et al. (2010) and Crowley et al. (2012). Attempts were made to estimate both of these models but convergence of the likelihood function (to a maximum value) could not be achieved. Humphreys et al. (2010) have pointed this out as a particular limitation of applying double hurdle models in particular. The inability to compare the results of the double hurdle model with the models proposed is a drawback of the study In addition, Jaunky and Ramchurn (2014) estimated both the log normal and truncated versions of the hurdle model and found the log normal model to be a better fit statistically.

5. In doing this we also assume that the error terms in the probit model and the expenditure model are not correlated with one another. This assumption can be tested by estimating the Heckman model. The coefficient on the Inverse Mills ratio in the Heckman model represents the covariance between the two error terms. In both the bookmaker/tote model and the lottery model this was found to be insignificant suggesting that independence exists between the error terms.

6. Our elasticities are estimated based on a semi-elasticity formula, i.e. (dy/dx)*(1/y), because our explanatory variables are either discrete or in logarithms. Thus we examine the proportionate change in y for a change in x rather than a proportionate change in x. Because of this, the estimated elasticities will be the same as the estimated coefficients. From Table 4

7. Time dummies were incrementally excluded if their p-value was greater than 0.10. Likelihood ratio tests support the use of the restricted final model over the unrestricted initial model (Bookmaker/tote first stage probit, LR χ2 (11) = 7.96, p = 0.717; Bookmaker/tote second stage OLS, LR χ2 (16) = 8.93, p = 0.916; Lottery first stage probit, LR χ2 (12) = 5.42, p = 0.943; Lottery second stage OLS, LR χ2 (13) = 9.00, p = 0.773).

8. Using these dummy variables as explanatory variables could introduce simultaneous bias or reverse causality in our model. The finding that the association between the two forms of gambling is stronger in one direction may, however, alleviate the potential for this problem.