We then estimate equations of the form:
where the coefficient of interest is now y. So we are now asking whether deviations in cigarette taxes from their state-specific mean cause a relative change in the happiness of predicted smokers relative to those unlikely to smoke. We also include the set of covariates, X, that were used to predict smoking, and which may have independent effects on happiness, as well as a full set of state and year dummies (and, for Canada, month dummies, since some of the Canadian GSS surveys are carried out throughout the year and we want to capture any seasonality in survey responses).
We create dummy variables for the various happiness responses described above, and use those as our dependent variables. We estimate linear probability models for ease of interpretation; probit estimates are similar. In estimating all our equations, we adjust the standard error to allow for both auto-correlation and the grouped data, as suggested by Bertrand, Duflo and Mullainathan. We do this by performing a White correction that allows for an arbitrary variance-covariance matrix within states. buy claritin online
Table 2 shows our basic findings. The first three columns focus on American Data, while the second three focus on Canadian data. Each regression is an OLS estimate of, including covariates, where the dependent variable is a dummy indicating which level of happiness people chose. Where the covariates used are the same in both countries, we use one row for both regressions; where they differ, we use separate rows for the U.S. and Canadian cases.
In the U.S. data, cigarette taxes have a positive but insignificant effects on the probability of predicted smokers answering “very happy” or “pretty happy”. It has a negative and very significant effect on the probability of answering “Not happy”. Specifically, the interaction term between the predicted smoking variable and the tax rate in column is significantly negative. This suggests that cigarette taxes especially reduce unhappiness amongst those predicted to be smokers. Our estimated effect here is that each penny of excise taxation reduces unhappiness by 0.156 percentage points among predicted smokers. Given that the effect is focused on reduction in unhappiness, we focus on this variable for the remainder of our U.S. analysis.
In the next three columns, we examine the effect of Canadian tax changes on happiness in the Canadian data. Strikingly, we once again find that higher cigarette taxes make predicted smokers happier. Taxes raise the probability of predicted smokers answering “very happy”, while reducing the probability of them answering “somewhat happy” or “unhappy”. Once again, the statistically most significant effects are found here for unhappiness, where we find that each cent of excise taxation lowers the odds of being unhappy by 0.048 percentage points. To parallel the U.S. analysis, we focus on this unhappiness measure for the rest of the paper.