Item nonresponse occurs inevitably in almost all surveys conducted by statistical agencies because some sampled units refuse to respond to sensitive items or may not know the answer to some items. Item nonresponse is usually treated by using single imputation, which consists of creating a single value to replace a missing value. The main effect of item nonresponse is that when the respondents and the nonrespondents are different with respect to the survey variables, then nonresponse bias is introduced. Also, since the observed sample size is smaller than the sample size initially planned, nonresponse has the effect of leading to estimators with larger variance than that which would have been obtained if complete response had been achieved. This increase in variance is called the nonresponse variance. Finally, imputation has the effect of distorting the relationships between variables. In this paper, using data observed in the context of the Canadian Community Health Survey, we propose to investigate empirically the properties of estimators of population means, domain means and finite population coefficients of correlation in terms of bias and mean square error when regression imputation has been used to fill in the missing values. The exposition is easily accessible to readers with some background in linear regression.