Consumer spending on entertainment and the Great Recession

ABSTRACT

This paper empirically investigates potential effects of economic recessions on consumers’ decision-making process for leisure activities using the Consumer Expenditure Survey data during the Great Recession. We employ the Probit model to study how changes in income affect the likelihood of making nonzero expenditures on entertainment activities. Recognizing the presence of a high degree of censoring, we employ the Tobit model to assess the income effect on recreational activities to avoid bias in the least squares estimator for the latent coefficients. Income coefficient estimates are significantly positive in all years we consider, confirming that leisure is a normal good. However, we observe statistically significant decreases in the income coefficient during recession years in two out of three categories of leisure activities. That is, the responsiveness of consumption to income changes decreases during recession years, which implies a sluggish adjustment in leisure expenditures when economic distress is elevated.

KEYWORDS:

• Consumer Expenditure Survey
• entertainment
• Great Recession
• probit
• tobit

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. One referee concerns about the existence of an issue of multicollinearity. Severe multicollinearity may cause inefficient estimates and/or unstable coefficient estimates. In what follows, our key model estimates are overall statistically significant. Further, as can be seen in our previous manuscript Gao, Kim, and Zhang (2015), which is based on separate estimations for 4 different years, coefficient estimates for most characteristic variables (other than the income) seem stable over time. So, we believe our models do not suffer from a severe multicollinearity problem.

2. The probability of each event is, $Pr\left(y_i=1\right)=Pr\left(y_i^{\ast }>0\right)=\mathrm{\Phi }\left(X_i^T\beta \right)=\mathrm{\Phi }\left({\beta }_0+{\beta }_1{X}_{i1}+{\beta }_2{X}_{i2}+\cdots +{\beta }_k{X}_{ik}\right)$ and $Pr\left(y_i=0\right)=Pr\left(y_i^{\ast }\le 0\right)=1-\mathrm{\Phi }\left(X_i^T\beta \right)=1-\mathrm{\Phi }\left({\beta }_0+{\beta }_1{X}_{i1}+{\beta }_2{X}_{i2}+\cdots +{\beta }_k{X}_{ik}\right)$, where $\mathrm{\Phi }\left(\cdot \right)$ is the Gaussian cumulative distribution function.

3. Marginal effect of $x_j$ is $\mathrm{\Phi }\left(X_i^T\beta \hspace{0.17em}\right)\frac{\mathrm{\partial }{X}_i^T\beta }{\mathrm{\partial }{x}_{ij}}=\mathrm{\Phi }\hspace{0.17em}\left(X_i^T\beta \right)\hspace{0.17em}{\beta }_j$. Since the marginal effect changes depending on the location of $i$, we report average marginal effects.

4. See Gao et al. (2015) for the OLS estimates.

5. One alternative explanation about the decrease in the intercept is that consumers increased their spending on entertainment-related equipment such as iPods and iPads which became very popular since the mid-2000s. Because our models do not include proxy variables for such technological innovations, those potentially positive effects on expenditures might have been included in the intercept, dominating negative effects from recessions.

6. Some measures of the goodness of fit such as McFadden Pseudo R² and Veall-Zimmermann Pseudo R² are available upon request, which range from 0.10 to 0.25. It should be noted that we use parsimonious models to focus on the income effect during recession years, so the goodness of fit is not our major concern. That is, we are primarily interested in statistically meaningful changes of the income effect.

7. We do not report biased OLS estimates to save space. For OLS results, see Gao et al. (2015).

8. One referee suggests implementing a similar analysis for consumption of non-entertainment goods or services to rigorously show if such reallocation occurs during recession years. We agree with this suggestion but we leave it to a future study because the topic is beyond the scope of this manuscript.