Explaining State-to-State Differences in Seat Belt Use: A Multivariate Analysis of Cultural Variables David W. Eby & Lisa J. Molnar TRB ANB45 Occupant Protection Committee Meeting July 26, 2011

Acknowledgements • Co-Investigators: • Yang Yang, Kohinoor Dasgupta, Vijayan N. Nair (U-M Statistics) • Stephen Pollock (U-M Engineering) • Sponsored through a cooperative agreement between U-M and the National Highway Traffic Safety Administration (DTNH22-06-H-00055)

Background • Overall US belt use: 85% • Belt use among states varies from 68% (Wyoming) to 98% (Michigan). • There also is regional variation in belt use (e.g., belt use tends to be higher in Western states). • Many factors are known to (or could) influence a state’s belt use: • Policy factors (secondary vs. primary law; fine amounts) • Demographic/environmental factors (gender; age; race; SES; vehicle type; obesity; urbanicity)

• Most often, these factors are usually examined in isolation.

Background • Goal: Use a multivariate approach to gain a better understanding of the socio-demographic variables that influence statewide belt use rates. • Examine the influence of statewide “cultural” variables that might relate to belt use. • The idea that culture can influence traffic safety behaviors is not new. The AAAFTS published a set of papers on this topic in 2007.

Methods • Find a suitable dataset • Self-reported data (MVOSS) were examined and discarded because there were too few respondents in most states. • Direct observation data (NOPUS) were examined and discarded because there were no state identifiers available. • Crash data (2008 FARS) were examined and found to be suitable for the project. • We only considered drivers of passenger vehicles and cases in which belt use/nonuse was indicated (N=19,090). • Cases were from all states, excluding Hawaii and DC.

Percent Deviation in Belt Use from National Average

• Verify that state-to-state differences in belt use through FARS analysis

Red: Orange: White: Light Grn: Green:

<-10% -10% to <-5% -5% to +5% >+5% to +10% >+10%

Methods

• Analyses then explored the extent that known factors also influenced belt use in FARS. Variable Gender Vehicle Type Urbanicity Law Type Fine Amount

Female Male Automobile Pickup Rural Urban Primary Secondary <= $30 > $30

Belt Use (%)

Number of Drivers

75.4 62.3

5,651 13,439

69.1 60.3

12,742 6,348

60.7 73.8 58.3 69.7 64.3 76.1

11,133 7,957 5,911 13,179 15,998 3,092

Distributions of Seat Belt Use Percentages and Number of Drivers by Categories (Binary Variables Only)

Methods • Belt use also varied by age and BMI (continuous variables).

Left Panel: Proportion of Statewide Seat Belt Use by Driver’s Age. Right Panel: Proportion of Statewide Seat Belt Use by Driver’s BMI

Preliminary Model • Fitted the following logistic regression model to the driver-level seat belt use data:

Regression Coefficient Baseline Urbanicity Fine Law Gender Vehicle Age BMI

-0.66 0.56 0.51 0.40 0.54 0.26 0.02 -0.01

Std. Error

z-value

p-value

0.03 0.05 0.03 0.04 0.03 0.00 0.00

17.02 10.86 11.83 14.28 7.45 14.75 -1.92

0.00 0.00 0.00 0.00 0.00 0.00 0.06

Differences among Statewide Belt Use Rates: Observed minus Predicted Preliminary Model

Preliminary Model

Red: Orange: White: Light Grn: Green:

<-10% -10% to <-5% -5% to +5% >+5% to +10% >+10%

“Cultural” Variables

• The factors in the preliminary model did not predict all the variation in state-to-state belt use rates. • Variables had to be available state-by-state. • Potential cultural categories: • Socio-demographics (known factors); • Education (% age >25 with a HS degree or more); • Race (% White); • Income (Median household income); • Political Leaning (% pop who voted democratic in 2008 National presidential election) • Religiosity (% of pop who said yes to question: “Is religion important to your daily life?”)

Statewide Seat Belt Use by Percent Democratic and Republican

Statewide Seat Belt Use by Education and Median Income

Statewide Seat Belt Use by Religiosity and Percent White.

Full Model Regression Coefficient

Baseline Urbanicity Fine Law Gender Vehicle Age Religiosity White Democratic Education Income

1.02 0.45 0.54 0.40 0.57 0.21 0.02 -0.03 -0.01 0.02 -0.01 0.00

Std. Error

z value

p-value

0.03 0.05 0.05 0.04 0.04 0.00 0.00 0.00 0.01 0.01 0.00

13.17 10.45 7.96 15.08 6.02 14.92 -7.14 -3.24 3.15 -1.28 1.40

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20 0.16

Selection and Validation of Final Model • Because of correlations among these variables, a systematic variable selection technique was used (AIC). • Variables were systematically dropped and differences in predicted outcomes were compared (AIC differences). • Little difference between the full model and a model that did not include education and income. • A similar result was found when the AIC analysis was conducted separately on a 70/30 random split of the FARS data.

Final Model Regression Coefficient Baseline Urbanicity Fine Law Gender Vehicle Age Religion White Democratic

1.03 0.46 0.54 0.36 0.59 0.22 0.02 -0.03 -0.01 0.03

Std. Error

z-value

p-value

0.04 0.06 0.06 0.05 0.04 0.00 0.00 0.00 0.01

11.33 8.89 6.22 12.92 5.34 12.92 -6.80 -6.04 2.86

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Differences among Statewide Belt Use Rates: Observed minus Predicted Final Model

Red: Orange: White: Light Grn: Green:

<-10% -10% to <-5% -5% to +5% >+5% to +10% >+10%

Conclusions • Cultural factors seem to play a large role in belt use behavior at the state level. • The cultural variables of religiosity and political leaning help us to explain some of the differences in state-to-state variation in belt use. • The variables we investigated did not explain all differences among state belt use rates. • More detailed research is needed to better understand the relationships among religiosity and political leaning, and belt use. • Different constructs for these factors; • More years of FARS data; • Examine final model using direct observation data.