The first thing we show, in column (1) of Table 3, is that the results are not simply the spurious correlation of high Internet usage late in the sample with the large rise in prices due to the tobacco settlement. We do this by restricting the sample to only the years before fiscal year 1999 (when the settlement raised pre-tax prices substantially in our data). The results are almost exactly the same.
Second, in column (2), we exclude the four states with the lowest cigarette taxes (VA, KY, NC, SC); because they are frequently the source of the Internet cigarettes, taxable sales in these states might conceivably respond quite differently to changes in Internet usage. They are only a small segment of the sample but, as would be expected, the point estimates do show slightly more sensitivity when they are excluded. Column (3) weights each observation by population, in case the results are being driven by a few outlying observations in small states. If anything, the estimated influence of Internet usage on the price elasticity of taxed sales is higher.
In columns (4) and (5) we consider the role of Native American reservations as an alternative source of smuggling. As detailed in Evans, et al. (2002), a loosening of the rules regarding gambling on reservation land in 1989 caused a dramatic increase in the number of Indian casinos in the United States in the last 15 years. To the extent that more people are going to such casinos and are, then, able to pick up cigarettes tax free when they are there, this will make the price sensitivity of sales in a state more sensitive to tax rates due to tax avoidance, but will be only spuriously correlated with the growth of Internet-related tax avoidance over the same time period.
We have no direct measures of the number of Indian gaming visits by state across time. However, using the data from the National Indian Gaming Association, we have been able to count the number of Indian casinos by state in 2004, and in these two columns we split the sample into states where there were no casinos in 2004 (column 4) and states in the highest quartile that had more than five casinos by 2004 (column 5). The results show that while the baseline price elasticity is greater in the no-casino states, there is no difference in the impact of the Internet on the increase in that elasticity. It is the identical magnitude in both cases. We found the same thing interacting with the number of casinos in the state rather than splitting the sample.
In column (6) we deal with the issues of changing demographics. For example, if there has been a rise in teen smoking, and teens are both relatively price-sensitive (as documented in Gruber, 2000) and tend to live in states where the Internet grew fastest, this could cause us to spuriously conclude that rising Internet use makes taxed cigarette sales more tax-sensitive. We suspect, though, that any measure of the actual change of demographic characteristics in a state will almost certainly not give an explanation as to why the tax sensitivity of cigarette sales has nearly doubled in such a short time frame. The changes and the differences in elasticities across groups are simply too small. To further investigate this possibility, or any other state-specific factor, we allow the baseline elasticity to differ in every state (in addition to the existing state and year dummies accounting for differing levels of consumption). The specification also includes the Internet interaction term, so it examines whether higher Internet use makes states more price-sensitive than they would otherwise have been and accounts for any state-level differences in price sensitivity. In this specification the estimated impact of the Internet on tax sensitivity is statistically significant and still large, though smaller than in the baseline specification. The average t-statistic on the state level price elasticities is almost seven, so there is still enough variation to estimate these separately.
In column (7) we examine whether the Internet effect on the price elasticity can be differentiated from a linear trend in the elasticity that applies to all states. We do this by adding an explanatory variable that interacts the log of price with the year. This exercise suggest that, indeed, there is an upward drift in the (absolute value of) the price elasticity, but that this trend can be statistically differentiated from the impact of Internet use on the price elasticity. In column (7) the Internet effect is large, although not as large as in the baseline specifications.
Finally, in columns (8) and (9), we take the robustness check to the full extreme—in (8) allowing a linear time trend and state-specific elasticities, and in (9) allowing every state and every year to have a separate baseline elasticity—and identify the impact of the Internet on price sensitivity relative to these. Here the results break down. The point estimate of the interaction term is not significant and, in column (9), the point estimate has the wrong sign. Further, none of the state or year price elasticities are individually significant, either (average t-statistics of only around 0.2). The data are simply unable to estimate all of these effects separately.
Second, in column (2), we exclude the four states with the lowest cigarette taxes (VA, KY, NC, SC); because they are frequently the source of the Internet cigarettes, taxable sales in these states might conceivably respond quite differently to changes in Internet usage. They are only a small segment of the sample but, as would be expected, the point estimates do show slightly more sensitivity when they are excluded. Column (3) weights each observation by population, in case the results are being driven by a few outlying observations in small states. If anything, the estimated influence of Internet usage on the price elasticity of taxed sales is higher.
In columns (4) and (5) we consider the role of Native American reservations as an alternative source of smuggling. As detailed in Evans, et al. (2002), a loosening of the rules regarding gambling on reservation land in 1989 caused a dramatic increase in the number of Indian casinos in the United States in the last 15 years. To the extent that more people are going to such casinos and are, then, able to pick up cigarettes tax free when they are there, this will make the price sensitivity of sales in a state more sensitive to tax rates due to tax avoidance, but will be only spuriously correlated with the growth of Internet-related tax avoidance over the same time period.
We have no direct measures of the number of Indian gaming visits by state across time. However, using the data from the National Indian Gaming Association, we have been able to count the number of Indian casinos by state in 2004, and in these two columns we split the sample into states where there were no casinos in 2004 (column 4) and states in the highest quartile that had more than five casinos by 2004 (column 5). The results show that while the baseline price elasticity is greater in the no-casino states, there is no difference in the impact of the Internet on the increase in that elasticity. It is the identical magnitude in both cases. We found the same thing interacting with the number of casinos in the state rather than splitting the sample.
In column (6) we deal with the issues of changing demographics. For example, if there has been a rise in teen smoking, and teens are both relatively price-sensitive (as documented in Gruber, 2000) and tend to live in states where the Internet grew fastest, this could cause us to spuriously conclude that rising Internet use makes taxed cigarette sales more tax-sensitive. We suspect, though, that any measure of the actual change of demographic characteristics in a state will almost certainly not give an explanation as to why the tax sensitivity of cigarette sales has nearly doubled in such a short time frame. The changes and the differences in elasticities across groups are simply too small. To further investigate this possibility, or any other state-specific factor, we allow the baseline elasticity to differ in every state (in addition to the existing state and year dummies accounting for differing levels of consumption). The specification also includes the Internet interaction term, so it examines whether higher Internet use makes states more price-sensitive than they would otherwise have been and accounts for any state-level differences in price sensitivity. In this specification the estimated impact of the Internet on tax sensitivity is statistically significant and still large, though smaller than in the baseline specification. The average t-statistic on the state level price elasticities is almost seven, so there is still enough variation to estimate these separately.
In column (7) we examine whether the Internet effect on the price elasticity can be differentiated from a linear trend in the elasticity that applies to all states. We do this by adding an explanatory variable that interacts the log of price with the year. This exercise suggest that, indeed, there is an upward drift in the (absolute value of) the price elasticity, but that this trend can be statistically differentiated from the impact of Internet use on the price elasticity. In column (7) the Internet effect is large, although not as large as in the baseline specifications.
Finally, in columns (8) and (9), we take the robustness check to the full extreme—in (8) allowing a linear time trend and state-specific elasticities, and in (9) allowing every state and every year to have a separate baseline elasticity—and identify the impact of the Internet on price sensitivity relative to these. Here the results break down. The point estimate of the interaction term is not significant and, in column (9), the point estimate has the wrong sign. Further, none of the state or year price elasticities are individually significant, either (average t-statistics of only around 0.2). The data are simply unable to estimate all of these effects separately.
