Analysis strategyWe use several types of regression to assess the relationship between the dependent and independentvariables. First, we report linear regression estimates for the model that treats fear of online financial crimeas a function of background characteristics that indicate vulnerability, device possession and onlinevictimization experiences. Results were checked for normal distribution of residuals and homoscedasticity –two assumptions in linear regression modelling – and both were found to be in order. Secondly, the analysesthat treat online activities (purchasing and banking) as dependent variables employ two types of regressionmodelling: logistic (on whether or not respondents perform this activity) and negative binomial (on howmany hours per week, on average, respondents devote to it). Considering the highly skewed nature of thelatter variables (counts, with many respondents reporting few or an intermediate amount of hours, and asmall group devoting a considerable amount of time to it), negative binomial models are the most suitableoption (Osgood, 2000). Finally, the last set of regression analyses treat (dichotomous) PC security measurestaken as dependent variables; logistic regression is used here. All models were checked for multicollinearity,which was found to be not an issue considering that VIF values nowhere exceeded 1.5.