Potential sources of observer bias in observational studies of police

Very little research has focused on the potential for observer bias in observational studies of police because "...there are almost no systematic data with respect to observer effects in observation studies in natural settings" (Reiss 1968:358). The purpose of this paper is to synthesize fragmented accounts of observer bias in the field research literature by: (1) defining and describing four types of observer bias; (2) operationalizing one type of observer bias (reactivity) by identifying key independent and dependent variables; and (3) deriving hypotheses which can be tested using qualitative and quantitative data from a large scale observational study of police (Project on Policing Neighborhoods, or POPN).;An in-depth review of POPN qualitative data on reactivity for patrol officers revealed that reactivity is often embedded within social exchanges with observers which include: (1) social cues from patrol officers; or (2) explicit changes in officer behavior. POPN descriptive data was also converted into a coding scheme in order to document the prevalence of social cues which were potential sources of reactivity. One key finding from the qualitative analysis was that the level of reactivity within observational data (in the form of social cues or explicit changes in patrol officer behavior) depends on the specificity of the data.;Multivariate analyses were also conducted in order to determine if the qualitative coding (or selected themes derived from the qualitative data on reactivity) or other potential sources of reactivity derived from the field research literature (i.e., observer sex, status congruency, time in the field) were isolated instances or exerting a systematic effect on patrol officer behavior in the form of significant effects within multivariate equations. Multivariate analyses were performed at three different units of analysis and examined multiple aspects of patrol officer behavior. Encounter level multivariate analyses focused on patrol officer's decision to arrest and use of force. Ride level multivariate analyses looked at the level of aggressive patrol and amount of goofing off an officer engaged in per shift. Ride segment multivariate analyses focused on variation in patrol officer behavior within a ride. Selected findings from the multivariate analyses show that; (1) patrol officers who express concerns about safety are less likely to arrest suspects; (2) patrol officers are more likely to use force against citizens if the observer helps the police officer in some capacity over the course of the shift; and (3) patrol officers are less likely to use force against citizens if they are less familiar with the observer. In addition, patrol officer behavior during the first hour of the shift is significantly different than the rest of the shift.