In the previous three chapters we discussed several estimators of the prevalence of crime under the assumption that the missing data were missing at random. Here we examine the validity of this assumption and find that it is not supported by the data. HUs that have had at least one month of nonresponse in the year of interest have consistently lower cheery indicators and thus nonresponse and victimization are positively correlated. Because of this positive correlation, modelling of the missing data leads to lower estimates of θ.
Unable to display preview. Download preview PDF.