OBJECTIVES:   Proxy respondents are often used in multi-item national health care experience surveys when the desired respondents are unable to answer, including in surveys whose results are used for pay for performance. Missing items are often addressed through complete case analysis (CCA) or using the mean score if at least half of the scale items are completed (half scale imputation, “HSI”). However, item nonresponse may differ for proxies and patients. Different adjustment approaches may produce different results, with implications for payment. The research aims to determine if the choice of imputation method affects the impact of proxy reports on experience and quality scores.  METHODS:   We analyzed nationally representative cross-sectional survey data from patients (or their proxies) with incident lung or colorectal cancer. Linear regression models included an independent variable for proxy status (0/1, patient as reference group), and additional patient, site and clinical covariates. Four outcomes were evaluated, with higher responses reflecting better reports/ratings: medical care, nursing care, and care coordination (all multi-item scales, 0-100), and care quality rating (single item, 0-4). Three different imputation methods for missing outcome item data were compared: CCA, HSI, and multiple imputation (MI).  RESULTS:   Adjusted analyses revealed similar average proxy scores across imputation methods. The largest differences for proxy scores were seen for the medical care outcome: MI=+1.28 (SE 0.63) points, HSI=+1.57 (SE 0.63) points, CCA=+1.70 (SE 0.65) points. For care quality rating, the MI and CCA proxy scores were both statistically not significant.  CONCLUSIONS:   The impact of proxy reports on care experience and quality outcomes is minimally affected in this dataset by the imputation method for item nonresponse. The ability to impute covariates and include a larger number of participants under MI makes findings more representative, thus MI is recommended. Simulation analyses to identify when different methods may affect results are needed.