Search results

Prior analyses of racial bias in the New York City’s Stop-and-Frisk program implicitly assumed that potential bias of police officers did not vary by crime type and that their decision of which type of crime to report as the basis for the stop did not exhibit any bias. In this paper, we first extend the hit rates model to consider crime type heterogeneity in racial bias and police officer decisions of reported crime type. Second, we reevaluate the program while accounting for heterogeneity in bias along crime types and for the sample selection which may arise from conditioning on crime type. We present evidence that differences in biases across crime types are substantial and specification tests support incorporating corrections for selective crime reporting. However, the main findings on racial bias do not differ sharply once accounting for this choice-based selection.

This chapter examines the use of mathematical programming to remove systematic bias from demand forecasts. A debiasing methodology is developed and applied to demand data from an actual service operation. The accuracy of the proposed methodology is compared to the accuracy of a well-known approach that utilizes ordinary least squares regression. Results indicate that the proposed method outperforms the least squares approach.

This chapter analyzes the properties of an alternative least-squares based estimator for linear panel data models with general predetermined regressors. This approach uses backward means of regressors to approximate individual specific fixed effects (FE). The author analyzes sufficient conditions for this estimator to be asymptotically efficient, and argue that, in comparison with the FE estimator, the use of backward means leads to a non-trivial bias-variance tradeoff. The author complements theoretical analysis with an extensive Monte Carlo study, where the author finds that some of the currently available results for restricted AR(1) model cannot be easily generalized, and should be extrapolated with caution.

Control charts are designed to be effective in detecting a shift in the distribution of a process. Typically, these charts assume that the data for these processes follow an approximately normal distribution or some known distribution. However, if a data-generating process has a large proportion of zeros, that is, the data is intermittent, then traditional control charts may not adequately monitor these processes. The purpose of this study is to examine proposed control chart methods designed for monitoring a process with intermittent data to determine if they have a sufficiently small percentage of false out-of-control signals. Forecasting techniques for slow-moving/intermittent product demand have been extensively explored as intermittent data is common to operational management applications (Syntetos & Boylan, 2001, 2005, 2011; Willemain, Smart, & Schwarz, 2004). Extensions and modifications of traditional forecasting models have been proposed to model intermittent or slow-moving demand, including the associated trends, correlated demand, seasonality and other characteristics (Altay, Litteral, & Rudisill, 2012). Croston’s (1972) method and its adaptations have been among the principal procedures used in these applications. This paper proposes adapting Croston’s methodology to design control charts, similar to Exponentially Weighted Moving Average (EWMA) control charts, to be effective in monitoring processes with intermittent data. A simulation study is conducted to assess the performance of these proposed control charts by evaluating their Average Run Lengths (ARLs), or equivalently, their percent of false positive signals.

Data fusion and the combination of multiple data sources have been part of travel survey processes for some time. In the current context, where technologies and information systems spread and become more and more diverse, the transportation community is getting more and more interested in the potential of data fusion processes to help gather more complete datasets and help give additional utility to available data sources. Research is looking for ways to enhance the available information by using both various data collection methods and data from various sources, surveys or observation systems. Survey response rates are decreasing over the world, and combining survey modes appears to be an interesting way to address this problem. Letting interviewees choose their survey mode allows increasing response rates, but survey mode could impact the data collected. This paper first discusses issues rising when combining survey modes within the same survey and presents a method to merge the data coming from different survey modes, in order to consolidate the database. Then, it defines and describes the data fusion process and discusses how it can be relevant for transportation analysis and modelling purposes. Benefiting from the availability of various datasets from the Greater Montréal Area and the Greater Lyon Area, some applications of data fusion are constructed and/or reproduced to illustrate and test some of the methods described in the literature.

Research in the area of forecasting and stock inventory control for intermittent demand is designed to provide robust models for the underlying demand which appears at random, with some time periods having no demand at all. Croston’s method is a popular technique for these models and it uses two single exponential smoothing (SES) models which involve smoothing constants. A key issue is the choice of the values due to the sensitivity of the forecasts to changes in demand. Suggested selections of the smoothing constants include values between 0.1 and 0.3. Since an ARIMA model has been illustrated to be equivalent to SES, an optimal smoothing constant can be selected from the ARIMA model for SES. This chapter will conduct simulations to investigate whether using an optimal smoothing constant versus the suggested smoothing constant is important. Since SES is designed to be an adapted method, data are simulated which vary between slow and fast demand.

Economists and sociologists have proposed arguments for why there can exist wage penalties for work involving helping and caring for others, penalties borne disproportionately by women. Evidence on wage penalties is neither abundant nor compelling. We examine wage differentials associated with caring jobs using multiple years of Current Population Survey (CPS) earnings files matched to O*NET job descriptors that provide continuous measures of “assisting & caring” and “concern” for others across all occupations. This approach differs from prior studies that assume occupations either do or do not require a high level of caring. Cross-section and longitudinal analyses are used to examine wage differences associated with the level of caring, conditioned on worker, location, and job attributes. Wage level estimates suggest substantive caring penalties, particularly among men. Longitudinal estimates based on wage changes among job switchers indicate smaller wage penalties, our preferred estimate being a 2% wage penalty resulting from a one standard deviation increase in our caring index. We find little difference in caring wage gaps across the earnings distribution. Measuring mean levels of caring across the U.S. labor market over nearly thirty years, we find a steady upward trend, but overall changes are small and there is no evidence of convergence between women and men.

A regression kink design (RKD or RK design) can be used to identify casual effects in settings where the regressor of interest is a kinked function of an assignment variable. In this chapter, we apply an RKD approach to study the effect of unemployment benefits on the duration of joblessness in Austria, and discuss implementation issues that may arise in similar settings, including the use of bandwidth selection algorithms and bias-correction procedures. Although recent developments in nonparametric estimation (Calonico, Cattaneo, & Farrell, 2014; Imbens & Kalyanaraman, 2012) are sometimes interpreted by practitioners as pointing to a default estimation procedure, we show that in any given application different procedures may perform better or worse. In particular, Monte Carlo simulations based on data-generating processes that closely resemble the data from our application show that some asymptotically dominant procedures may actually perform worse than “sub-optimal” alternatives in a given empirical application.