Biased Coefficient Estimates Research Articles (Page 5)

Although much market-based accounting research is based on regressions of abnormal returns on contemporaneous unexpected earnings, many have despaired about the intrinsic ability of accounting earnings to explain stock returns. These regressions exhibit low R2, lower than expected coefficients on unexpected earnings (ERC's), and various unusual features including non-linearity, lower R2 and response coefficients for loss firms, and lower R2 and response coefficients for high-growth and high-tech firms. Some improvement in explanatory power has been achieved by including various proxies for information that is currently available about future period earnings. This paper contributes to that line of research by deriving a specification, from the abnormal earnings model, that extends the traditional ERC regression by including current period forecast revisions of future period earnings. Relative to the traditional regression, the full specification increases R2 substantially, reduces the bias in coefficient estimates (caused by omitted correlated variables), and mutes the three unusual features mentioned above.

The logistic regression procedure is a popular statistical method used when analyzing quantal dose-response data. However, logistic regression results based on a poorly designed experiment can be seriously compromised. Our results indicate that depending on the spacing of the doses, the number of doses, and the number of replications at each dose, the user can get very misleading results including ineffective lack-of-fit tests and severely biased coefficient estimates along with biased estimates of response. In addition variance formulas based on asymptotic theory may be completely inappropriate. Simulation results are used to support these statements.

We use surveys of households and health-care facilities conducted in the same area at the same time to determine which characteristics of providers attract users of contraceptives. By using the full-information maximum-likelihood technique to jointly estimate choice of contraceptive method and choice of provider, we avoid self-selection bias. Results support the need for modeling quality and for jointly estimating the choice of contraceptive method and the choice of provider to avoid biased estimates of coefficients. The results suggest that for the Cebu, Philippines region, small local clinics that focus on family planning tend to be most favored by clients.

Modeling accident frequencies as zero-altered probability processes: An empirical inquiry

This article reports empirical explorations of how well the predictive mean matching method for imputing missing data works for an often problematic variable—income—when income is used as an explanatory variable in a substantive regression model. It is found that the performance of the predictive mean method varies considerably with the predictive power of the imputation regression model and the percentage of cases with missing data on income. In comparisons of single-value with multiple-imputation methods, it also is found that the amount of bias and the loss of precision associated with single-value methods is considerably less than that associated with a weak imputation model. Situations in which using imputed data can lead to seriously biased estimates of regression coefficients (and related statistics) and situations in which the bias is so minimal as to be nonproblematic are identified.

Due to the unobservability of the new credit production, most of the empirical loan market studies use, instead, the observable credit stock. This substitution has been pointed out to be likely to generate biases (e.g. see Lown and Peristiani (1996)). In this paper, we show that under quite unrestrictive conditions, this substitution does not lead to biased estimates of any log-log model coefficients, as long as banks panel data is used and fixed effects are included in the estimated equation.

Statistical models of infrastructure facility deterioration are typically estimated using panel data sets of in-service facilities. For example, biannual ratings of bridges have been used to develop discrete models of component deterioration by a number of researchers. Unfortunately, these models have not accounted for the presence of heterogeneity in the panel data, which may lead to biased coefficient estimates. Furthermore, researchers have usually imposed a Markovian specification in the development of such models, implying that the probabilistic deterioration in a given period is independent of history. This assumption may be unrealistic for some types of facilities in which early distress initiation leads to accelerated deterioration in later stages of their lives. In this paper, we adopt a random-effects specification to control for heterogeneity in a probit model of bridge-deck deterioration and extend the model to investigate the presence of state dependence. The proposed model yields improved results in comparison with a simple probit model and provides evidence that is inconsistent with the Markovian assumption in bridge-deck deterioration. An implication of this study is that both heterogeneity and state dependence may need to be accounted for in developing probabilistic infrastructure deterioration models.

Statistical analyses of data from epidemiological studies of workers exposed to radiation have been based on recorded annual radiation doses. It is usually assumed that the annual doses are known exactly, although it is generally recognized that the data contain uncertainty due to measurement error and bias. We propose the use of a probability distribution to describe an individual's dose during a specific period and develop statistical methods for estimating this distribution. The methods take into account the "measurement error" that is produced by the dosimetry system and the bias that was introduced by policies of recording doses below a threshold as zero. The method is applied to a sample of dose histories over the period 1945 to 1955 obtained from hard-copy dosimetry records at Oak Ridge National Laboratory (ORNL). The result of this evaluation raises serious questions about the validity of the historical personnel dosimetry data that are currently being used in studies of the effects of low doses in nuclear industry workers. In particular, it appears that there was a systematic underestimation of doses for ORNL workers. This may result in biased estimates of dose-response coefficients and their standard errors.

A four‐equation model is used to investigate the effects of political instability (PI) on the savings rate in Sub‐Saharan Africa. Utilising a comprehensive measure of PI, we find that political instability has a deleterious effect on the savings rate both directly and indirectly through a reduction in investment and economic growth. The negative effects of PI on savings rate occurs contemporaneously as well as with a lag. We also find that economic growth has a stabilising effect on the political system and that not accounting for these effects through a simultaneous equations model results in biased coefficient estimates. These relationships are robust with respect to model specification. The implication of our results is that ‘economic factors’ alone cannot explain the development process in Less Developed Countries.

The National Voter Registration Act of 1993 mandates "motor voter" programs in all states prior to the 1996 presidential election. Using state-level registration and turnout data over the 1976-1992 period, this study finds that motor voter programs already implemented in many states have significantly increased participation rates. A duration-based specification of motor voter is employed, to account for the fact that driver's license renewal cycles last up to six years or even more in some states. Dummy-variable specifications are shown to underestimate the eventual impact of motor voter. Models include state dummy variables to control for long-standing differences in participation rates across states that otherwise bias coefficient estimates for registration closing date and other variables. In contrast to motor voter, other provisions required by the NVRA--including mail-in and agency-based registration, and limitations on the purging of voter rolls--show little evidence of effectiveness in the states where they have already been implemented.

Ordinary survival models implicitly assume that all individuals in a group have the same risk of death. It may, however, be relevant to consider the group as heterogeneous, i.e. a mixture of individuals with different risks. For example, after an operation each individual may have constant hazard of death. If risk factors are not included, the group shows decreasing hazard. This offers two fundamentally different interpretations of the same data. For instance, Weibull distributions with shape parameter less than 1 can be generated as mixtures of constant individual hazards. In a proportional hazards model, neglect of a subset of the important covariates leads to biased estimates of the other regression coefficients. Different choices of distributions for the unobserved covariates are discussed, including binary, gamma, inverse Gaussian and positive stable distributions, which show both qualitative and quantitative differences. For instance, the heterogeneity distribution can be either identifiable or unidentifiable. Both mathematical and interpretational consequences of the choice of distribution are considered. Heterogeneity can be evaluated by the variance of the logarithm of the mixture distribution. Examples include occupational mortality, myocardial infarction and diabetes.

Ordinary survival models implicitly assume that all individuals in a group have the same risk of death. It may, however, be relevant to consider the group as heterogeneous, i.e. a mixture of individuals with different risks. For example, after an operation each individual may have constant hazard of death. If risk factors are not included, the group shows decreasing hazard. This offers two fundamentally different interpretations of the same data. For instance, Weibull distributions with shape parameter less than 1 can be generated as mixtures of constant individual hazards. In a proportional hazards model, neglect of a subset of the important covariates leads to biased estimates of the other regression coefficients. Different choices of distributions for the unobserved covariates are discussed, including binary, gamma, inverse Gaussian and positive stable distributions, which show both qualitative and quantitative differences. For instance, the heterogeneity distribution can be either identifiable or unidentifiable. Both mathematical and interpretational consequences of the choice of distribution are considered. Heterogeneity can be evaluated by the variance of the logarithm of the mixture distribution. Examples include occupational mortality, myocardial infarction and diabetes.

The ordinary least squares (OLS) estimator gives biased coefficient estimates if coefficients are not constant for all cases but vary systematically with the explanatory variables. This article discusses several different ways to estimate models with systematically and randomly varying coefficients using estimated generalized least squares and maximum likelihood procedures. A Monte Carlo simulation of the different methods is presented to illustrate their use and to contrast their results to the biased results obtained with ordinary least squares. Several applications of the methods are discussed and one is presented in detail. The conclusion is that, in situations with variables coefficients, these methods offer relatively easy means for overcoming the problems.

Errors in recorded security prices are a source of misspecification in the market model. If recorded price errors are sufficiently nonrandom, they result in biased returns and in biased and inconsistent estimates of market model regression coefficients. This paper argues that tax-induced flow-supply pressures cause end-of-the-year recorded price errors to be nonrandom enough to create the appearance of anomalous turn-of-the-year stock return behavior. Empirical tests of returns and market model regression coefficients during the turn-of-the-year period cannot reject this errors-in-variables explanation of the turn-ofthe-year effect. The turn-of-the-year (TOY) effect (or January effect) refers to the anomal? ous behavior of stock returns during the last five trading days in December and the first five trading days in January. This anomaly is of particular interest to financial researchers because it appears to be a small-firm effect and the source of the majority of size-related anomalies (see Keim (1985), Reinganum (1981), and Roll (1983a), 1983b)). The interest in the TOY effect is justified because of its implications concerning the validity of the Capital Asset Pricing Model (CAPM) and market efficiency. In this paper, we show that there is a price-related effect operating during the TOY period. First, we show that the magnitude of the TOY return is inversely related to price. When we control for price, most of the size-related TOY effect found by Reinganum (1983) and Roll (1983a) disappears. Second, the data indicate that all firms have unusually high returns during the TOY period compared with their returns during the rest of the year. Third, we find seasonal shifts in the market model betas associated with the TOY period. Furthermore, we test the hypothesis that the TOY effect is an errors-in-variables problem due to the use of the one-eighth pricing convention in recording security prices and the accommodation of tax-induced flow pressures by liquidity traders (such as market

in this paper we specify and estimate a structural limited dependent variable model with which we study both the health and retirement status of the elderly. Standard linear estimators, which assume that these variable sare continuous, are not appropriate and categorical estimation techniques are preferred. Our model differs from previous work in that we have longitudinal data and random effects that are correlated over time for different individuals. The problem is made more complicated because there is sample truncation, which could potentially bias coefficient estimates, since approximately twenty percent of the individuals in our sample die. We outline the full information maximum likelihood estimator for such a model and implement it in our empirical analysis. With our structural estimates we analyze, among other things, the degree to which endogeneously determined health status affects the probability of retirement and how changes in social security benefits and eligibility for transfer payments modify both healthiness and the demand for leisure.

Recent studies have documented a significant rise in the male black-white earnings ratio since the mid-1960s. The growing difference in nonemployment rates of blacks and whites clouds these optimistic findings. The basic question addressed in this paper is whether selectivity bias, caused by racial differences in employment rates, is a serious problem in the estimation of wage functions for adult males. For males age 21-34 we found no evidence of selectivity bias, but for the older cohort of males age 35-54, the results are quite different. For both whites and blacks, there is strong positive selection bias. It appears that biased estimates of several important coefficients are obtained using simple ordinary least squares procedures. The most interesting of these are the effect of low education for blacks.

Specification error in multinomial logit models