We analyze the results of a national survey collected in two modes: self-administered instrument on the web with personal phone interview follow-up of web non-respondents.We apply regression and implied utility-multiple imputation mode effect adjustments. Since some items may exhibit mode effects, such as social desirability bias, a split-sample design has been built into the study, with 13% of the cases randomized into phone-only condition. Such randomization allows for a rigorous comparison of the item response distributions in the two modes. We analyze the behavioral and attitude items to identify the ones that may have been affected by the mode effect. A logistic model for Yes/No responses or an ordinal logistic model for Likert scales was fit to the data with explanatory variables that included demographic variables and the mode indicator for the subsample of the mode compliers. The regression mode effect adjustments consists of zeroing out the mode variable when forming the predictions based on the estimated regressions, and can be extended to the entire sample. Another mode adjustment is based on econometric framework of implied utilities in logistic regression modeling. We simulated implied utilities of the different responses, followed up by selection of the response with the greatest utility. This is essentially an imputation procedure for the response in the less reliable mode (phone personal interview), and requires the framework of multiple imputation to obtain reliable standard errors. The variables that exhibited the strongest mode effects were found to be the self-reported incidence of donating one’s time to family and neighbors (possible over-reporting due to social desirability bias) andmajor financial problems in the last 5 years (possible under-reporting due to
social desirability bias). The standard errors of the adjusted estimates have gone up, as expected.