MINORITY THREAT AND POLICE STRENGTH FROM 1980 TO 2000: A FIXED‐EFFECTS ANALYSIS OF NONLINEAR AND INTERACTIVE EFFECTS IN LARGE U.S. CITIES

• Date: 01 February 2014
• Published date: 01 February 2014
• DOI: http://doi.org/10.1111/1745-9125.12027
• Author: DAVID JACOBS

CORRECTION

MINORITY THREAT AND POLICE STRENGTH FROM

1980 TO 2000: A FIXED-EFFECTS ANALYSIS OF

NONLINEAR AND INTERACTIVE EFFECTS IN LARGE

U.S. CITIES∗

DAVID JACOBS

Department of Sociology, Ohio State University

In 2005, Stephanie L. Kent and I published an article in this journal on the relationship

between racial threat and police strength in large U.S. cities (Kent and Jacobs, 2005).

The results were based on two-way fixed-effects pooled time-series analysis of these cities

in 1980, 1990, and 2000. When another graduate student recently reexamined this data,

he found a serous coding error and a few less critical variable discrepancies that may

be attributable to reporting agency data modifications, which occurred after these data

originally were collected. This reanalysis shows the results when these differences and a

more serious error are rectified.

Fixed-effects estimation only captures the influence of change in variables. This feature

is extremely valuable because the influence of any omitted but pertinent explanatory vari-

able that does not vary automatically is removed in fixed-effects models. Bias attributable

to any time-invariant omitted variables therefore is eliminated. Yet the value of an ex-

planatory variable must change in at least one or more cases in order for a variable to be

included in a fixed-effects model. In the original publication (Kent and Jacobs, 2005), we

entered a dummy variable coded “1” if a city had a city manager. The fixed-effects algo-

rithm did not remove this time-invariant explanatory variable because coding mistakes

indicated this variable had different within-case values.

Allison (2009) provided a remedy by showing how an analysis largely conducted with

fixed effects nevertheless can estimate the influence of explanatory variables even if the

variables in question are time invariant. To use this method, one must subtract each case

mean from each time-varying explanatory variable and enter these mean-centered vari-

ables together with their means as explanatory variables in a model. If time-invariant

explanatory variables are entered and if such a model is estimated using random effects,

the coefficients on the mean-centered explanatory variables will be equivalent to those

estimated by fixed effects. The coefficients on the time-invariant explanatory variables,

however, will be estimated with random effects. To provide a correct estimate of the ef-

fects of city manager presence, I report here a reanalysis using Allison’s method.

Table 1 shows the corrected results using the variables in the best model in the

prior publication (Kent and Jacobs, 2005; see model 5 in table 5). With but two minor

∗Direct all correspondence to David Jacobs, Department of Sociology, Ohio State University, 238

Townshend Hall, 1885 Neil Avenue, Columbus, OH 43210-1222 (e-mail: Jacobs.184@osu.edu).

C2014 American Society of Criminology doi: 10.1111/1745-9125.12027

CRIMINOLOGY Volume 52 Number 1 140–142 2014 140