More recently, probability surveys are facing with challenges of falling response rates, rising costs, and coverage of frames. These concerns have led to an increased interest in the use of nonprobability samples. However, in the use of no matter probability or non-probability samples, the estimates can be biased if the samples deviate from the target populations of interest. To reduce bias in estimates, auxiliary data especially administrative records (ARs) can be used in an effort to deal with selection and other biases. However, confidentiality concerns have hampered the use of ARs, especially when continuous variables are collected in ARs. We proposed a response propensity prediction (RPP) approach that allows best use of continuous auxiliary variables commonly measured in surveys and ARs while protecting confidentiality in individual-level data. We applied the RPP approach to estimate mean depression score of soldiers serving in Ohio Army National Guard (OANG) in 2008-2009 using a sample of 2,616 soldiers in the OANG Mental Health Initiative and the AR of 12,570 soldiers in the population.