Pei-Fen Kuo

Examining the effects of site selection criteria for evaluating the effectiveness of traffic safety countermeasures

The primary objective of this paper is to describe how site selection effects can influence the safety effectiveness of treatments. More specifically, the goal is to quantify the bias for the safety effectiveness of a treatment as a function of different entry criteria as well as other factors associated with crash data, and propose a new method to minimize this bias when a control group is not available. The study objective was accomplished using simulated data. The proposed method documented in this paper was compared to the four most common types of before-after studies: the Naïve, using a control group (CG), the empirical Bayes (EB) method based on the method of moment (EB(MM)), and the EB method based on a control group (EB(CG)). Five scenarios were examined: a direct comparison of the methods, different dispersion parameter values of the Negative Binomial model, different sample sizes, different values of the index of safety effectiveness (θ), and different levels of uncertainty associated with the index. Based on the simulated scenarios (also supported theoretically), the study results showed that higher entry criteria, larger values of the safety effectiveness, and smaller dispersion parameter values will cause a larger selection bias. Furthermore, among all methods evaluated, the Naïve and the EB(MM) methods are both significantly affected by the selection bias. Using a control group, or the EB(CG), can mutually eliminate the site selection bias, as long as the characteristics of the control group (truncated data for the CG method or the non-truncated sample population for the EB(CG) method) are exactly the same as for the treatment group. In practice, finding datasets for the control group with the exact same characteristics as for the treatment group may not always be feasible. To overcome this problem, the method proposed in this study can be used to adjust the Naïve estimator of the index of safety effectiveness, even when the mean and dispersion parameter are not properly estimated.

Estimating safety performance trends over time for treatments at intersections in Florida

Researchers have put great efforts in quantifying Crash Modification Factors (CMFs) for diversified treatment types. In the Highway Safety Manual (HSM), CMFs have been identified to predict safety effectiveness of converting a stop-controlled to a signal-controlled intersection (signalization) and installing Red Light Running Cameras (RLCs). Previous studies showed that both signalization and adding RLCs reduced angle crashes but increased rear-end crashes. However, some studies showed that CMFs varied over time after the treatment was implemented. Thus, the objective of this study is to investigate trends of CMFs for the signalization and adding RLCs over time. CMFs for the two treatments were measured in each month and 90-day moving windows respectively. The ARMA time series model was applied to predict trends of CMFs over time based on monthly variations in CMFs. The results of the signalization show that the CMFs for rear-end crashes were lower at the early phase after the signalization but gradually increased from the 9th month. On the other hand, the CMFs for angle crashes were higher at the early phase after adding RLCs but decreased after the 9th month and then became stable. It was also found that the CMFs for total and fatal/injury crashes after adding RLCs in the first 18 months were significantly greater than the CMFs in the following 18 months. This indicates that there was a lag effect of the treatments on safety performance. The results of the ARMA model show that the model can better predict trends of the CMFs for the signalization and adding RLCs when the CMFs are calculated in 90-day moving windows compared to the CMFs calculated in each month. In particular, the ARMA model predicted a significant safety effect of the signalization on reducing angle and left-turn crashes in the long term. Thus, it is recommended that the safety effects of the treatment be assessed using the ARMA model based on trends of CMFs in the long term after the implementation of the treatment.

Comparison of Application of Product of Baseline Models and Accident-Modification Factors and Models with Covariates: Predicted Mean Values and Variance

Research was done to compare the application of full models (or models with several covariates) and baseline models, estimated by using data meeting specific nominal conditions combined with accident-modification factors (AMFs) for predicting motor vehicle crashes. The analysis focuses on the predicted values and associated inferences for both types of models. In the past few years, researchers have questioned the approach of multiplying baseline models with AMFs. For comparison, full and baseline models are estimated by using data collected on rural four-lane highways in Texas. AMFs describing the safety effects related to lane and shoulder width as well as the number of horizontal curves per mile are extracted from previous work. Two scenarios describing AMF values and their associated uncertainty are evaluated. The results of the study show that the product of baseline models and AMFs produces a much larger variance, hence a wider 95% predicted confidence interval, than the variance calculated using the full models. This is consistent for the two scenarios evaluated and for all levels of uncertainty associated with the AMFs. The study concludes that the full model should be used instead of the product of the baseline model and AMFs when the study objective includes variance as part of the decision-making process.

Estimating the safety impacts in before–after studies using the Naïve Adjustment Method

ABSTRACT The before–after study is the most popular approach for estimating the safety impacts of an intervention or treatment. Recent research, however, has shown that the most common before–after approaches can still provide a biased estimate when an entry criterion is used and when the characteristics of the treatment and control groups are dissimilar. Recently, a new simple method, referred to as the Naïve Adjustment Method (NAM), has been proposed to mitigate the limitations identified above. Unfortunately, the effectiveness of the NAM using ‘real’ data has not yet been properly investigated. Hence, this paper examined the accuracy of the NAM when the treatment group contains sites that have different mean values. Simulated and two observed datasets were used. The results show that the NAM outperforms the Naïve, the Control Group, and the empirical Bayesian methods. Furthermore, it can be used as a simpler alternative for adjusting the Naïve estimators documented in previous studies.

Modeling the Spatial Effects on Demand Estimation of Americans with Disabilities Act Paratransit Services

A reliable method for predicting paratransit ridership is important, especially for the efficiency of the services offered. The commonly used aggregate regression model is most accurate for forecasting the total demand for regional areas such as whole counties or cities; however, it is likely to be geographically inaccurate. This paper proposes a geographical weight regression (GWR) model for predicting the demand for the types of para-transit services required by the Americans with Disabilities Act. The GWR model reflects better the characteristic of each area having its own coefficient for predictors rather than the same value throughout. The results show that trip demand increased proportionately to (a) the population size, (b) the ratio of senior citizens, (c) the ratio of people below the poverty line, and (d) the ratio of African-American riders. These results suggest that the predictive performance of the GWR model is better than that of the ordinary least squares (OLS) regression model. The GWR model is of greater value than the OLS model to researchers and practitioners, because the predictor variables are readily available from census data; this availability of data allows researchers to use the model after calibration.

Variable Selection of Travel Demand Models for Paratransit Service: A Data Mining Approach

In order to forecast the demands for American Disability Act (ADA) travel, many complicated factors are needed to be considered, including, but not limited to socioeconomic data and service operational characteristics. In other words, the choice of suitable explanatory variables to construct an ADA travel demand model is an important and complex task. Data mining techniques provide a promising way to select the explanatory variables. They work especially well in situations where the number of relevant variables is large and where the interactions among variables or models are not clear. In this study, the authors applied data mining techniques for selecting variables and building models. Census data was mined to select candidate variables. Also, they compared the performances of three types of models: (1) a traditional linear model, (2) a traditional model with variables selected by the classification and regression tree (CART) method, and (3) a traditional model with the variables selected by the random forest (RF) method. The results show that the fraction of senior citizens (age > 65), average household size (owner occupied), fraction of African Americans, fraction of Hispanics, annual household income, fraction of males, median age, fraction of households with a family member with a disability, and the total population, are the significant variables for modeling ADA travel demands.