Effectiveness of Different Policies Aimed to Reduce Traffic Deaths

This project aims to explore the effectiveness of different policies in reducing traffic deaths in the United States. The data used in this project covers 336 observations from 48 states over seven years (1982-1988).

The policies studied include beertax, jaild, comserd, and mlda. Several control variables were considered, including the state unemployment rate, per capita personal income, GSP rate of change, and others.

The data was analyzed using graphical representations and a pooled OLS model with cluster robust standard errors. A fixed effects model was also used to control for unobserved heterogeneity.

The results show that only per capita personal income, per capita consumption of pure alcohol, and the Mormon population were significant in the pooled model. However, after controlling for unobserved heterogeneity, jaild and comserd became significant, indicating that these policies may be effective in reducing traffic deaths.

Why?

• This project examines policies that could effectively reduce alcohol-involved vehicle fatalities, which is a major public health concern.

• The analysis uses data from 48 states and covers a period of seven years, making it a good generalization for the US population.

• Provides insight into the effectiveness of policies such as beer tax, mandatory jail sentence, mandatory community service, and minimum legal drinking age. It also highlights the need to consider unobserved heterogeneity when analyzing panel data.

Overview

Data Set Description

The data are for the “lower 48” U.S. states (excluding Alaska and Hawaii), annually for 1982 through 1988. The traffic fatality rate is the number of traffic deaths in a given state in a given year, per 10,000 people living in that state in that year. Traffic fatality data were obtained from the U.S. Department of Transportation Fatal Accident Reporting System. The beer tax is the tax on a case of beer, which is an available measure of state alcohol taxes more generally. The drinking age variables are binary variables indicating whether the legal drinking age is 18, 19, or 20. The two binary punishment variables describe the state’s minimum sentencing requirements for an initial drunk driving conviction: “Mandatory jail?” equals one if the state requires jail time and equals zero otherwise, and “Mandatory community service?” equals one if the state requires community service and equals zero otherwise. Total vehicle miles traveled annually by state was obtained from the Department of Transportation. Personal income was obtained from the U.S. Bureau of Economic Analysis, and the unemployment rate was obtained from the U.S. Bureau of Labor Statistics.

Data Exploration

First, we organized and tried to comprehend the data which is based on 336 observations (N) from 48 States (Entities) through 7 time periods of years 1982-1988 (T).

We wanted to study what policies effectively reduce mraidall (alcohol-involved VFR), which is referenced as Vehicle Fatality Rate X 10,000 (VFR) from this point on. The policies we focused on while studying this data were: beertax (tax on case of beer in dollars), jaild (mandatory jail sentence), comserd (mandatory community service) and mlda (minimum legal drinking age).

Table 1.1 shows variables that we initially thought could be significant in our model with some transformed into logarithmic form for better readability and interpretation.

We began with the following control variables:

·        State unemployment rate (%): unrate

·        Log per capita personal income ($): lnperinc

·        GSP rate of change: gspch

·        % of drivers aged 15-24: yngdrv

·        % residing in dry counties: dry

·        Log Population: lnpop

·        Log ave. mile per driver: lnvmiles

·        Log population of 15-17 year olds: lnpop1517

·        Log population of 18-20 year olds: lnpop1820

·        Log population of 21-24 year olds: lnpop2124

·        Per capita pure alcohol consumption (annual, gallons): spircons

·        % Southern Baptist: sobapt

·        % Mormon: mormon

Image 1.1 shows evidence that our panel data is balanced.

Table 1.2 displays the summary for all the variables mentioned above and we found that the mean VFR is 0.6593 and mean beertax was $0.5132 per case of beer. On average, ~28% of the of the 48 continuous states had mandatory jail sentences and ~18.5% of the states had mandatory community service.

Next, we used graphical representations to interpret the policies and some relevant variables for insight on the data.

Graph 1.1 displays the mean mraidall (VFR) with respect to each state. The states with noticeable changes included Mississippi, Montana, New Hampshire, Texas, Wyoming, and South Carolina. The VFR in Mississippi experienced a sharp decline in the first 2 years followed by a sharp increase before it catches a constant rate. Montana had a bumpy decreasing trend as well as New Hampshire and Wyoming. Texas had a smooth decreasing trend from 1982 to 1988. South Carolina though, had an increasing trend for four years before decreasing in the end.

Graphs 1.2- 1.5 imply that during the period where unemployment rate decreased, the VFR also decreased. When the logarithmic values of personal income increased, the VFR decreased. We expected that with the increase on beer tax, VFR would go down; however, this was not the case. VFR decreased even when beer tax decreased, suggesting that beer tax is not an effective policy.

Graphs 1.6 – 1.7 imply that mandatory jail sentences and mandatory community service helped reduce the VFR. However, when VFR peaks in 1986, so did mandatory jail time and community service, which implies that these two variables may not be the most effective policies.

Graph 1.8-1.10 shows that if the beer tax is above the mean, then VFR was higher on average and had a fluctuating trend. If the beer tax is below the mean, then VFR was lower overall. If the minimum legal drinking age is 18, VFR had an upward trend over the years. At 19, VFR decreased before spiking up in 1987. For age 20, VFR decreased before bouncing up in 1984 and then increasing sharply in 1986 onwards. For age 21, it was an overall decrease throughout the years with a more stable pattern compared to the others.

We ran a pooled OLS model with Cluster Robust standard errors that included all variables we thought would affect our dependent variable VFR. According to the results, we found that none of the variables were significant at the 5% level except lnperinc, spircons and mormon.

According to pooled model when other factors are kept constant, if per capita personal income increases by 1%, we expect VFR to decrease by approximately 0.00844 and a 1-gallon annual increase in the per capita consumption of pure alcohol is expected to increase the VFR by 0.062. A 1% increase in Mormon population is expected to decrease the VFR by 0.005.

This model though is not applicable to our data since there is an endogeneity problem.

We used the fixed effects model to control for unobserved heterogeneity with the same variables from the pooled model. The following are findings from the results compared to the pooled OLS:

• Magnitude for beertax has increased but is still insignificant even at the 10% significance level.

• Magnitude for jaild increased by more than twice its value from the pooled model. The coefficient for comserd increased by almost five times. These variables have now become statistically significant at the 5% level.

• If all factors are kept constant, a mandatory jail sentence is expected to increase VFR by 0.213 compared to if there was no jail time and mandatory community service is expected to decrease VFR by 0.2 compared to if there was no mandatory community service.

• Minimum legal drinking age is still insignificant.

• Spircons is even more significant, with its coefficient increasing by ~4 times. A 1-gallon annual increase in the per capita consumption of pure alcohol is expected to increase the VFR by 0.264.

• The magnitude for unrate is almost the same; however, it is now significant at the 10% level.

• lnperinc now has decreased by almost half and is not statistically significant anymore.

• The rest of the variables remain insignificant.

Next, we did an entity and time fixed effects model since omitted variables like the development and presence of safer cars might vary over time but not across economic entities. The results showed that jaild, spircons and comserd are significant at the 5% level but the rest are not. Even though beertax and mlda are not significant, we kept them in the model since they are the policies we are monitoring.

States with mandatory jail sentences are expected to have a higher VFR by 0.233 compared to states without and states with mandatory community service are expected to have a 0.22 lower VFR compared to those without, given all other variables are constant. A 1-gallon annual increase in the per capita consumption of pure alcohol is expected to increase the VFR by 0.311.

Then, we tested the significance of our time effects to see if they are jointly statistically significant. According to the test, we have a p-value of 0.0021 signifying that at least one year influences the model, so we continued to use the time variables.

We regressed a time fixed effect models without these insignificant variables: gspch, dry, pop, and vmiles (restricted model 1). We saw that jaild, comserd, unrate and spircons are statistically significant. A mandatory jail sentence is expected to increase VFR by 0.24 more than if there was no jail time and mandatory community service is expected to decrease VFR by 0.21 more than if there was no mandatory community service, given all other variables are constant. A 1-gallon annual increase in the per capita consumption of pure alcohol is expected to increase the VFR by 0.299 and a 1% increase in unemployment rate is expected to decrease VFR by 0.03.

The coefficients for the different population ages and yngdrv are only insignificant at the 10% level so they were removed from the model and we proceeded with our new time fixed effect model (restricted model 2).

In this model, we have variables beertax, jaild, comserd, mlda, unrate and spircons. At the 5% level, jaild, comserd, spircons and unrate are still significant but beertax and mlda are not. The coefficients for this model have not changed much from restricted model 1. The time variables also show that each year compared to 1982 had a reduction in VFR that was statistically significant at the 5% level except 1986 at the 10% level.

Moving forward, we took restricted model 2 variables from a time fixed effects and fitted them into a random effects model to see which one is better for our data and which variables explain our dependent variable well.

In our random effects model, we finally saw beertax become significant but only at the 10% level. Jaild is significant at the 5% level but comserd and unrate are only significant at the 10% level while mlda is still insignificant.

Mandatory jail sentence is expected to increase VFR by 0.205 and mandatory community service is expected to decrease VFR by 0.15, given all other variables are constant. Under the same conditions, a 1% increase in unemployment rate is expected to decrease VFR by 0.0256.

In the last part of our regression analysis, we performed a Hausman Test to determine which model to choose – fixed (entity and time) or random – with the following explanatory variables: beertax, jaild, comserd, mlda, unrate, spircons and time variables. We rejected the null hypothesis that there is no endogeneity and selected the entity time fixed effects model.

Regression Analysis

Summary and Conclusion:

After analyzing the models taken to observe how different factors and policies affected the alcohol-involved vehicle fatality rate, we concluded that the entity time fixed effects model was the most suitable approach (restricted model 2). This model controls for omitted variable bias and observed/unobserved heterogeneity, and is not randomly sampled data; therefore, the estimators will be unbiased and consistent.

The controlling variables that significantly impacted the VFR were per capita pure alcohol consumption and the unemployment rate. As alcohol consumption in a state increased annually by 1-gallon, the VFR increased by 0.264. A 1% increase in the unemployment rate resulted in a 0.032 decrease in VFR. While these factors are not policies, they can influence how states make policies to reduce alcohol-involved VFR.

All time indicator variables were significant at the 5% level except 1986 which was still significant at the 10% level. We saw that the VFR decreased each year when compared to 1982.

The beertax and minimum legal drinking age policies did not show to have a significant impact on the alcohol-involved VFR on this data. The policies that did significantly impact the alcohol-involved VFR were mandatory jail sentence and mandatory community service. States with a jail sentence had higher VFR on average than states without by 0.236; therefore, implementing a mandatory jail sentence is not an effective policy to reduce the alcohol-induced VFR. Having a mandatory community service though reduced the VFR by 0.195, so more states should consider implementing this policy.

*** For full code and outputs, check out the GitHub link up top***