HLDI Bulletin
|
Vol 33, No. 36 : December 2016 4
Vehicles
e study vehicles were the 10 most recent model years for each calendar year during 2009–10 and 2012–14. For ex-
ample, data from calendar year 2009 included model years 2001–10, whereas data from calendar year 2014 included
model years 2006–15. Total exposure and claims are shown in Table 2.
Table 2: Exposure and claims by coverage type
Coverage
Exposure (insured vehicle
years) Claims
Collision 1,4 67,3 37 83,298
Property damage liability 1,4 67,3 37 42,510
Bodily injury liability 1,506,548 7,2 6 5
Medical payment 1,481,167 10,199
Analysis methods
Claim frequency
Claim frequency was analyzed for 2009–10 when the road test in New Hampshire was mandatory, and for 2012–14 af-
ter the repeal. A Poisson regression logarithmic link function was used for each insurance coverage type to examine
the dierence in claim frequency for drivers 75 and older between New Hampshire and the comparison states using
drivers 55–74 as the comparison age group. e regression also controlled for a variety of covariates. e estimate for
the 2012–14 period was used as a baseline and subtracted from the 2009–10 estimate.
e main independent variables in the analysis include:
State: A categorical variable for the study state of New Hampshire and the two control states, Vermont and Maine.
Rated driver age: Rated driver ages were classied into two age groups: a study group of ages 75 and older, and a com-
parison group of ages 55–74. e control age group was selected because of enough loss data. It also spanned the typical
retirement ages of 65–67, allowing for the adjustment of a possible lifestyle change aer retirement.
Interaction of state and rated driver age: is categorical variable was designed to capture the dierent patterns in
claim frequencies between the study group in New Hampshire and the comparison age group in the control states.
e 55–74 age group in the comparison states served as the baseline.
Renewal cycle, in-person renewal frequency, and vision proof did not change during the study periods and thus were
not included in the model. Covariates included vehicle age, vehicle size and class, rated driver gender, rated driver
marital status, deductible range (collision coverage only), risk, vehicle density, and calendar year. e reference cat-
egories for the categorical independent variables were assigned to the values with the highest exposure: vehicle age =
6, vehicle size and class = midsize four-door cars, gender = female, marital status = married, risk = standard, deduct-
ible = $250-$499 (collision only), vehicle density = 250–499 vehicles per square mile, and calendar year = 2010 (with
road test, 2013 (without road test).
For space reasons, illustrative full regression results on collision claim frequency for 2009–10 are shown in Appendix
A. To further simplify the presentation, the exponent of the parameter estimate was calculated, 1 was subtracted, and
the result multiplied by 100. e resulting number corresponds to the eect of the feature on that loss measure. For
example, the estimate of collision claim frequency for older drivers subject to the road test was -0.0139, thus older
drivers in New Hampshire had a collision claim frequency 1.4 percent lower than in control states ((exp(-0.0139)
-1)*100 =-1.4).