Abstract
Understanding the major determinants of crash fatalities continues to be an important topic of investigation for safety researchers. Regression models using a vast number of explanatory variables are often used which result in a huge array of specifications. Results often vary among studies based on size of estimated coefficients and significance levels. To address this, we explore both significance and model sturdiness in regression models using Leamer’s s-values. This Bayesian technique allows us to address estimation uncertainty and model ambiguity over all possible subset regressions so as to evaluate the effect of key variables which we focus on as contributors to crash fatalities. These include cell phone use, fleet modernization, suicidal behavior, alcohol use, and speed limits.
Similar content being viewed by others
Notes
See NHTSA (2016).
See NHTSA (2016).
See Loeb et al. (1994) for a more complete list and discussion of these contributing influences.
See Blattenberger et al. (2013).
See for example, Fowles et al. (2010).
Strangely, the bans do not include hands-free devices even though research indicates that such devices have a similar adverse effect. See for example, Consiglio et al. (2003).
See also Fowles et al. (2013).
A time trend was also considered as a potential regressor in the model. However, our measure of the age of the fleet (MCMY) was found to be almost perfectly correlated with a time trend, with a correlation coefficient of 0.99962.
The target of interest in a Bayesian regression setting is the posterior means for the \( \beta \) parameters. Mathematical developments for EBA are found in Leamer (1982).
The Bayesian natural conjugate model sets the prior variance for the β’s = var(β) = v2Ik×k where Ik×k is the k by k identity matrix. Bounds are obtained via the scalar v2 which is set to the minimum or maximum expected R-square divided by k, see Leamer (2014, 2016) for details. Calculations are performed in the software R (R Development Core 2016). Also see Fowles (1988).
A super pessimist prior is to exclude a variable from a regression, so the prior mean is at zero and the variance is zero as well (prior R-square zero). In this paper, we do not consider this kind of strict prior.
Our method of imputing cell phone subscriptions correlates with the actual data with a correlation coefficient of 0.9943.
The per se law refers to legislation that makes it illegal to drive a vehicle at a blood alcohol level at or above the specified BAC level. BAC is measured in grams per deciliter.
See Office of the Federal Register (1980).
See Blattenberger et al. (2009).
Regional dummies were included as explanatory variables, but results are not shown in Table 2.
See Leamer (2014).
See Leamer (2014).
Leamer’s decision rule in the selection of models requires the absolute values of s-values to be greater than 1 and the absolute value of t statistics for the full OLS model to be greater than 2. Conformity between the sign of t simple and t all is an additional factor in model selection although not necessarily required. However, when signs differ between these two values, additional information should be considered.
It is important to note that all three cell phone variables are sturdy regardless as to one’s prior whether pessimistic, optimistic, or not having a preconceived notion at all. The results in Table 2 regarding cell phone subscriptions reveal an overall positive effect of cell phones on fatality rates, but the marginal effect is diminishing. (A pictorial representation of this is available from the authors.) One should remember that the nonlinear specification is the prior proposed by previous studies.
Again, regional dummy variables are included in the analysis, but not shown in Fig. 3.
For t statistics and s-values, the correlation is .89.
All variables are standardized to mean 0 and variance 1 so as to allow for unit-free comparisons.
See Breiman (2001).
See Blattenberger et al. (2013).
See Centers for Disease Control and Prevention, National Center for Injury Prevention and Control (2012).
These are Region 9 states in our model.
See Conner et al. (2001).
A note is also due regarding the potential shortcomings in the paper. Clearly, alternative models, e.g., double log models among others, could be investigated. One’s initial prior establishes the structure of the model, and s-values can be viewed then to demonstrate whether they appear reasonable. Quality of data is another matter in which all empirical investigations are subject to. One is left with using the best available at the time, recognizing measurement, and sampling errors are prone to show up. The data used in this study are obtained from well-known sources (as indicated in “Appendix 1”) and are sensitive to such errors no more than in other studies.
References
Beede KE, Kass SJ (2006) Engrossed in conversation: the impact of cell phones on simulated driving performance. Accid Anal Prev 38:415–421
Blattenberger G, Fowles R, Loeb PD, Clarke WA (2009) Understanding the cell phone effect on motor vehicle fatalities using classical and Bayesian methods. Presented at the Allied social sciences associations meeting (TPUG), San Francisco
Blattenberger G, Fowles R, Loeb PD, Clarke WA (2012) Understanding the cell phone effect on vehicle fatalities: a Bayesian view. Appl Econ 44:1823–1835
Blattenberger G, Fowles R, Loeb PD (2013) Determinants of motor vehicle crash fatalities using bayesian model selection methods. Res Transp Econ 43:212–222 (Special issue on: the economics of transportation safety)
Breiman L (2001) Statistical modeling: the two cultures. Stat Sci 16(3):199–231
Centers for Disease Control and Prevention, National Center for Injury Prevention and Control. Web-based Injury Statistics Query and Reporting System (WISQARS). http://www.cdc.gov/injury/wisqars/. Accessed 7 July 2012
Chaloupka FJ, Saffer H, Grossman M (1993) Alcohol-control policies and motor vehicle fatalities. J Legal Stud 22:161–186
Chapman S, Schoefield WN (1998) Lifesavers and samaritans: emergency use of cellular (mobile) phones in Australia. Accid Anal Prev 30:815–819
Conner KR, Cox C, Duberstein PR, Tian L, Nisbet PA, Conwell Y (2001) Violence, alcohol, and completed suicide: a case-control study. Am J Psychiatry 158:1701–1705
Connolly JF, Cullen A, McTigue O (1995) Single road traffic deaths: Accident or suicide? Crisis J Crisis Interv Suicide Prev 16(2):85–89
Consiglio W, Driscoll P, Witte M, Berg WP (2003) Effect of cellular telephone conversations and other potential interference on reaction time in braking responses. Accid Anal Prev 35:495–500
CTIA—The Wireless Association (2011). http://www.ctia.org. Accessed 10 Feb 2011
CTIA—The Wireless Association (2016). http://www.ctia.org/industry-data/ctia-annual-wireless-industry-survey. Accessed 7 Nov 2016
Etzerdorfer E (1995) Single road traffic deaths: Accidents or suicide? comment. Crisis J Crisis Interv Suicide Prev 16(4):188–189
Fowles R (1988) Micro EBA. Am Stat 4:274
Fowles R, Loeb PD (1992) The interactive effect of alcohol and altitude on traffic fatalities. South Econ J 59:108–112
Fowles R, Loeb PD (2019) Motor vehicle fatalities from the perspective of sturdy values: the autonomous vehicle effect. Presented at the American economic association and transportation and public utilities group meetings in Atlanta, GA, January 6, 2019
Fowles R, Loeb PD, Clarke WA (2010) The cell phone effect on motor vehicle fatality rates: a Bayesian and classical econometric evaluation. Transp Res Part E 46:1140–1147
Fowles R, Loeb PD, Clarke WA (2013) The cell phone effect on truck accidents: a specification error approach. Transp Res Part E. https://doi.org/10.1016/j.tre.2012.10.002
Freeborn BA, McManus B (2007) Substance abuse treatment and motor vehicle fatalities. College of William and Mary, Department of Economics, Working Paper Number 66
Gelman A (2008) Scaling regression coefficients by dividing by two standard deviations. Stat Med 27:2865–2873
Glassbrenner D (2005) Driver cell phone use in 2005—overall results. Traffic safety facts: research note, NHTSA, DOT HS 809967. http://www-nrd.nhtsa.dot.gov/Pubs/809967.PDF. Accessed 11 Feb 2011
Hoerl AE, Kennard RW (1970) Ridge regression: biased estimation for nonorthogonal problems. Technometrics 12(1):55–67
Huffine CL (1971) Equivocal single-auto traffic fatalities. Life Threat Behav 1(2):83–95
Laberge-Nadeau C, Maag U, Bellavance F, Lapiere SD, Desjardins D, Messier S, Saidi A (2003) Wireless telephones and risk of road crashes. Accid Anal Prev 35:649–660
Leamer EE (1978) Specification searches: ad hoc inference with non-experimental data. Wiley, New York
Leamer EE (1982) Sets of posterior means with bounded variance priors. Econometrica 50(3):725–736
Leamer EE (1983) Let’s take the con out of econometrics. Am Econ Rev 73(1):31–43
Leamer EE (2014) S-values and all subsets regressions. University of California at Los Angeles, working paper
Leamer EE (2016) S-values and Bayesian weighted all-subsets regressions. Eur Econ Rev 81:15–31
Loeb PD, Clarke W (2009) The cell phone effect on pedestrian fatalities. Transp Res Part E 45:284–290
Loeb PD, Talley WK, Zlatoper T (1994) Causes and deterrents of transportation accidents: an analysis by mode. Quorum Books, Westport
Loeb PD, Clarke WA, Anderson R (2009) The impact of cell phones on motor vehicle fatalities. Appl Econ 41:2905–2914
McEvoy SP, Stevenson MR, McCartt AT, Woodward M, Haworth C, Palamara P, Cercarelli R (2005) Role of mobile phones in motor vehicle crashes resulting in hospital attendance: a Case-Crossover Study. BMJ 33:428–435
Murray D, DeLeo D (2007) Suicidal behavior by motor vehicle collision. Traffic Injury Prev 8:244–247
Neyens DM, Boyle LN (2007) The effect of distractions on the crash types of teenage drivers. Accid Anal Prev 39:206–212
NHTSA (2016) Traffic safety facts 2016, DOT HS 812554, May 2018. https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812554. Accessed 28 Mar 2018
Office of the Federal Register (1980) National archives and records service, general service administration. The United States government manual: 1980–1981
Peltzman S (1975) The effect of automobile regulation. J Polit Econ 93:677–725
Phillips DP (1977) Motor vehicle fatalities increase just after publicized suicide stories. Science 196:1464–1466
Phillips DP (1979) Suicide, motor vehicle fatalities, and the mass media: evidence toward a theory of suggestion. Am J Sociol 84(5):1150–1174
Pokorny AD, Smith JP, Finch JR (1972) Vehicular suicides. Life Threat Behav 2(2):105–119
Porterfield AL (1960) Traffic fatalities, suicide, and homicide. Am Sociol Rev 25(6):897–901
Poysti L, Rajalin S, Summala H (2005) Factors influencing the use of cellular (mobile) phones during driving and hazards while using it. Accid Anal Prev 37:47–51
R Development Core (2016) R: a language and environment for statistical computing. http://www.R-project.org. Accessed 10 Jan 2016
Ramsey JB (1974) Classical model selection through specification error tests. In: Zarembka P (ed) Frontiers in econometrics. Academic Press, New York, pp 13–47
Ramsey JB, Zarembka P (1971) Specification error tests and the alternative functional form of the aggregate production function. J Am Stat Assoc Appl Sect 57:471–477
Redelmeier DA, Tibshirani RJ (1997) Association between cellular-telephone calls and motor vehicle collisions. N Engl J Med 336:453–458
Simmons WO, Welki A, Zlatoper TJ (2016) The impact of driving knowledge on motor vehicle fatalities. J Transp Res Forum 55(1):17–27
Souetre E (1988) Completed suicides and traffic accidents: longitudinal analysis in france. Acta Psychiatr Scand 77(5):530–534
Sullman MJ, Baas PH (2004) Mobile phone use amongst New Zealand drivers. Transp Res Part F Traffic Psychol Behav 7:95–105
Violanti JM (1998) Cellular phones and fatal traffic collisions. Accid Anal Prev 30:519–524
Welki A, Zlatoper TJ (2014) The effect of cell phones on international motor vehicle fatality rates: a panel-data analysis. Transp Res Part E Log Transp Rev 64:103–109
Welki A, Zlatoper TJ (2017) An analysis of illicit drug use and motor vehicle fatalities using contiguous state-level data. J Transp Res Forum 56(2):5–20
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix 1: Data sources
Name | Data source |
---|---|
FATAL | Highway Statistics (various years), Federal Highway Administration, Traffic Safety Facts (various years), National Highway Traffic Safety Administration |
MCMY | National Automobile Dealers Association (various years) and the National Household Travel Survey, US Department of Transportation. |
PERSE | Digest of State Alcohol-Highway Safety Related Legislation (various years), Traffic Laws Annotated 1979, Alcohol and Highway Safety Laws: A National Overview 1980, National Highway Traffic Safety Administration |
SPEEDHI | Highway Statistics (various years), Federal Highway Administration |
BELT | Traffic Safety Facts (various years), National Highway and Traffic Safety Administration |
BEER | US Census Bureau, National Institute on Alcohol Abuse and Alcoholism |
MLDA | A Digest of State Alcohol-Highway Safety Related Legislation (various years), Traffic Laws Annotated 1979, Alcohol and Highway Safety Laws: A National Overview of 1980, National Highway Traffic Safety Administration, US Census Bureau |
YOUNG | State Population Estimates (various years), US Census Bureau http://www.census.gov/population/www/estimates/statepop.html |
CELLPOP | Cellular Telecommunication and Internet Association Wireless Industry Survey, International Association for the Wireless Telecommunications Industry |
UNEMPLOY | Statistical Abstract of the United States (various years), US Census Bureau |
REALINC | State Personal Income (various years), Bureau of Economic Analysis website http://www.bea.doc.gov/bea/regional/spi/dpcpi.htm |
EDHS | Digest of Education Statistics (various years), National Center for Education Statistics, Educational Attainment in the United States (various years), US Census Bureau |
EDCOLL | Digest of Education Statistics (various years), National Center for Education Statistics, Educational Attainment in the United States (various years), US Census Bureau |
SUICIDE | Statistical Abstract of the United States (various years), US Census Bureau |
REGION | US States 1: ME, NH, VT; 2: MA, RI, CT; 3: NY, NJ, PA; 4: OH, IN, IL, MI, WI, MN, IA, MO; 5: ND, SD, NE, KS; 6: DE, MD, DC, VA, WV; 7: NC, SC, GA, FL; 8: KY, TN, AL, MS, AR, LA, OK, TX; 9: MT, ID, WY, CO, NM, AZ, UT, NV; 10: WA, OR, CA; 11: AK, HI |
Appendix 2: Descriptive statistics (primary variables, raw data)
Name | Mean | Median | Standard deviation |
---|---|---|---|
MCMY | 1987.00 | 1987.01 | 10.10 |
CELLPOP | 360.27 | 206.56 | 378.44 |
SPEED | 65.06 | 65.00 | 6.65 |
PERSE | 0.91 | 1.00 | 0.29 |
BEER | 1.29 | 1.28 | 0.23 |
MLDA | 20.74 | 21.00 | 0.76 |
BELT | 0.75 | 1.00 | 0.43 |
EDHS | 0.60 | 0.60 | 0.06 |
EDCOLL | 0.16 | 0.15 | 0.05 |
REALINC | 28,743.65 | 28,466.36 | 7227.34 |
YOUNG | 0.20 | 0.19 | 0.03 |
SUICIDE | 13.02 | 1.00 | 0.29 |
Rights and permissions
About this article
Cite this article
Fowles, R., Loeb, P.D. A sturdy values analysis of motor vehicle fatalities. Empir Econ 60, 2063–2081 (2021). https://doi.org/10.1007/s00181-020-01826-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-020-01826-2