Driving Safety versus State Alcohol Consumption and State GDP Per Capita

by robekulick

When I saw that Allstate publishes a rank of driver safety and also publishes data on average years between accidents for a given person in a number of cities in the United States, my natural inclination was to wonder how alcohol consumption and GDP by state relate to these measures.

Data:

http://www.allstatenewsroom.com/channels/News-Releases/releases/seventh-annual-allstate-america-s-best-drivers-report-reveals-safest-driving-cities

So I threw down my econ text book for an hour and went to work. The results are pretty interesting.

Here is the first regression I did, comparing the allstate rank to per capita alcohol consumption by state, per capita GDP by state, and geographic region:

Linear regression                                      Number of obs =     193
                                                       F(  6,   186) =   24.92
                                                       Prob > F      =  0.0000
                                                       R-squared     =  0.2351
                                                       Root MSE      =   49.43

------------------------------------------------------------------------------
             |               Robust
        rank |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     alcohol |   3.522978   12.44198     0.28   0.777    -21.02257    28.06853
         gdp |   .0014079   .0005037     2.80   0.006     .0004142    .0024015
        east |   5.043339   13.26603     0.38   0.704    -21.12789    31.21457
     midwest |  -79.25639   13.47801    -5.88   0.000    -105.8458   -52.66697
       south |  -48.17839   12.53827    -3.84   0.000     -72.9139   -23.44289
        west |  -54.58543   11.63552    -4.69   0.000    -77.53999   -31.63086
       _cons |   78.76122   34.10407     2.31   0.022     11.48071    146.0417
------------------------------------------------------------------------------

For the second I replaced the rank variable with the accident years variable:

 
Linear regression                                      Number of obs =     193
                                                       F(  6,   186) =   16.71
                                                       Prob > F      =  0.0000
                                                       R-squared     =  0.2931
                                                       Root MSE      =  1.4048

------------------------------------------------------------------------------
             |               Robust
      ayears |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     alcohol |    .013753   .3724981     0.04   0.971    -.7211113    .7486173
         gdp |  -.0000536   .0000125    -4.29   0.000    -.0000782   -.0000289
     midwest |    2.56402   .3887997     6.59   0.000     1.796996    3.331044
       south |   1.662039   .3538296     4.70   0.000     .9640036    2.360074
        west |   1.959656   .3447658     5.68   0.000     1.279502     2.63981
       other |   2.187849   .3539057     6.18   0.000     1.489664    2.886034
       _cons |   9.690933   .9040911    10.72   0.000     7.907342    11.47452

For the third I used the natural log of the accident years, alcohol, and gdp variables:

Linear regression                                      Number of obs =     193
                                                       F(  6,   186) =   23.70
                                                       Prob > F      =  0.0000
                                                       R-squared     =  0.3532
                                                       Root MSE      =  .15124

------------------------------------------------------------------------------
             |               Robust
    lnayears |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
   lnalcohol |  -.0844827   .0940677    -0.90   0.370    -.2700594     .101094
       lngdp |  -.4183175   .0642201    -6.51   0.000     -.545011   -.2916241
     midwest |   .2745044   .0504288     5.44   0.000     .1750186    .3739903
       south |   .1704175    .047285     3.60   0.000     .0771336    .2637014
        west |    .231661   .0467611     4.95   0.000     .1394107    .3239113
       other |   .3002159   .0535485     5.61   0.000     .1945755    .4058564
       _cons |   6.533847   .6644993     9.83   0.000     5.222923    7.844771
------------------------------------------------------------------------------

So what does this all mean? Well interestingly enough the effect of alcohol is not statistically significant in any of the regressions. Furthermore, the relationship between driving safety and GDP is statistically significant, but the two variables are inversely related. So drivers are less safe in wealthier states. Finally, and this won’t surprise anyone, drivers in the Midwest, South, West, and Alaska and Hawaii (my other category) are safer than drivers on the East Coast on average.

Now at this point I need to say something very, very, very important. THIS ABSOLUTELY DOES NOT INDICATE DRINKING AND DRIVING IS SAFE. We know for a fact it is not. But what this does mean is that some characteristic of states with higher levels of alcohol consumption is preventing this increase in translating into more traffic accidents. One possibility is that though people in some states drink more, they are no more likely to drive while drinking. Another possibility is that states with high alcohol consumption have stricter laws that reduce traffic accidents.

I’m not sure what to make of the fact that per capita GDP is inversely related to driving safety. I have to admit my thought was that after controlling for regional variation, the relationship would be positive.

And of course, anyone who has spent a significant amount of time on the east coast knows drivers here are crazy. One interesting question though is whether the Allstate measures effectively control for all of the effects of increased traffic on accidents. My initial thought was that by measuring things in terms of average time between accidents, the effects of higher traffic should be implicitly controlled for. But that probably calls for more research.

Advertisements