Data Description
The present paper conducts an econometric analysis of the article “Drunk Driving Legislation and Traffic Fatalities: New Evidence on BAC 08 Laws” in order to understand the data and methodology used. It is necessary to define the data and its features with which the researchers work in the report. The unit of observation in the article is the relative rates of fatalities per capita in each of the 48 states considered. In contrast, the unit of analysis is the phenomenon of adaptation of a number of laws regulating drunk driving (Freeman, 2007).
Investigators used two data points, 1982 and 2004, as benchmarks when the primary law BAC 08 was not implemented in any state and 47, respectively (Freeman, 2007). The dependent variable is traffic fatalities, which are divided into daytime and weekend nights per 100,000 population at the state level over the years 1980–2004 for the 48 contiguous states, taken from the FARS system and also from several affiliated researchers (Freeman, 2007). The impact of this indicator is assessed through the main independent variables of the BAC, ALR, GDL, seat belt, and speed limit laws, as well as demographics, traveled mileage, and business cycle (Freeman, 2007). Data were collected from 1980 to 2004 but used from 1982 onwards for the immediate analysis point.
Descriptive Statistics
Descriptive statistics are presented in the work in the corresponding table before the methodology. The element contains only a few columns – year, mean, standard deviation, minimum, and maximum, indicating the last two states where these indicators were achieved (Freeman, 2007). The first part of the table reflects continuous variables, and the second indicator only depicts the number of states that implemented specific laws and restrictions in the corresponding years.
The appendix contains information about effective dates for many listed regulations. Otherwise, regarding the median, separate information by state, which could be helpful for future research, was not provided, which is the only lack of information content of this aspect since the laws under discussion are not federal. Moreover, the dynamics could be reflected locally and on a large scale at different rates, which was analyzed much later (Scherer & Fell, 2019). In addition, it would have been worthwhile to indicate which data were dependent and which were independent variables rather than just separating the data by type. Therefore, the descriptive statistics are satisfactory but could have been better presented regarding variables and be more informative by implementing state-related details.
Econometric Techniques Employed
Furthermore, it is meaningful to explain the econometric techniques used in the report. In particular, a pooled cross-section approach was used, entangling time series analysis. The method is represented through a two-way fixed effects specification. It involves such parameters as an annual fatality rate, state-fixed and year-time effects, and vectors of indicator and control variables. Notably, the article also provides the application of alternative specifications pooled cross-section appraisals with updated common errors and event studies (Freeman, 2007). Therefore, the examined report utilizes one economic technique, and it is a pooled cross-section.
Regression Specification and Model Justification
Researchers use a comprehensive approach to assessment in the examined article. First, the work proposes an estimate of the static effect of the presence of laws using the above-mentioned pooled time series cross-section regressions (Freeman, 2007). Such an approach involves independent samples reckoned at two or more time points, which is often employed in socioeconomic studies that evaluate aspects in retrospect (Ferrari et al., 2021). Although the regression equation that gauges the static effect through the vector of inclusion of laws and other restrictions has an additional error term, investigators have agreed that it is necessary to estimate the lagged, not the static effect (Freeman, 2007).
A further approach was introduced with the LDV model, where a prediction error was introduced, providing a representation of the time lag through the event window. By subtracting these errors from the actual data, new values were obtained, which were more dynamic when taking into account the consequences of these laws. The calculations were made using Monte Carlo simulations and so-called pseudo-laws generated for specific sample years for each state.
Regression specifications, manifested through the criteria for including or excluding certain variables from the model, were carried out based mainly on economic aspects. First, the demographic data reflected in the descriptive statistics table included the percentage of youth and state unemployment rates. Such selection was guided by previous work and the frequent inclusion of this type of social data in examining community-relevant events to inform future research (Saeed et al., 2020).
Second, the differentiation of the dependent variable into weekend night fatalities and daytime is dictated by similar conditions for collecting data in the database from which they were extracted. Functional form issues are discussed in the context of an equation where the relevant effects are included, and the control and indicator variables are in the form of vectors (Freeman, 2007). The researchers did not conduct a preliminary analysis to determine such aspects for the variables – the information was taken in the form in which it is stored in open sources.
Assumptions and Estimation Validity
The investigators focus on some assumptions in this work. First, as discussed above, special attention is paid to the time lag that occurs when laws are adopted and their impact on the dependent variable. To solve this problem, the LDV model technique is used to estimate prediction errors and their variance (Freeman, 2007). Consistency manifests in introducing boundaries to evaluate error patterns for statistical significance.
Secondly, the selected pool of laws, which is considered indicator variables, is far from a complete list. For example, in California, over a certain period, about fifty similar legislative acts and measures were introduced that could affect the dependent variable (Freeman, 2007). Here, researchers already assume that the lack of opportunity to evaluate a specific law does not make it ineffective but only inseparable from the general context and number of regulations (Freeman, 2007). In such a case, the error term, state fixed effect, and time effect are implemented into the regression equation, but their values cannot always be calculated in principle. Accordingly, the assumptions about the data can be considered appropriate when necessary modifications are made to the estimation technique.
The introduction of the LDV approach for time lag adaptation significantly reduced the statistical significance of many of the results. In the methodology, the researchers did not describe in detail several econometric approaches that were reflected in the results. For example, they were considering this model without time series information and running regressions in a similar case (Freeman, 2007). Due to such assessments, which may indeed be essential in answering the research question but were not presented in the appropriate sections, the perception and interpretation of such information becomes more difficult.
Evaluation of Econometric Results
Although econometric approaches, in general, have been used quite widely to reveal the problem, their implementation and motivation in some cases are not clearly described by researchers. Moreover, a rather complex LDV modeling approach where boundaries were used solely to approximate statistical significance testing, although yielding high model performance results due to the adjusted R square metric, had significantly fewer statistically significant results than the static model (Freeman, 2007). As a result, in addition to the indicated shortcomings, the work leaves a good and understandable impression of the research conducted, where the connection between the introduction of laws and the reduction of fatalities in the states under consideration is visible.
References
Ferrari, G., Dulgheroff, P. T., Claro, R. M., Rezende, L. F., & Azeredo, C. M. (2021). Socioeconomic inequalities in physical activity in Brazil: A pooled cross-sectional analysis from 2013 to 2019. International Journal for Equity in Health, 20, 1-9. Web.
Freeman, D. G. (2007). Drunk driving legislation and traffic fatalities: New evidence on BAC 08 laws. Contemporary Economic Policy, 25(3), 293-308. Web.
Saeed, T. U., Nateghi, R., Hall, T., & Waldorf, B. S. (2020). Statistical analysis of area‐wide alcohol‐related driving crashes: A spatial econometric approach. Geographical Analysis, 52(3), 394-417. Web.
Scherer, M., & Fell, J. C. (2019). Effectiveness of lowering the blood alcohol concentration (BAC) limit for driving from 0.10 to 0.08 grams per deciliter in the United States. Traffic Injury Prevention, 20(1), 1-8. Web.