A Method of Estimation of Missing Values in Multivariate Data Suitable for Use with an Electronic Computer

Abstract

Summary Estimation of statistical parameters from multivariate data results in wasted information, if units with incomplete data are rejected entirely, and perhaps in inconsistencies in the variance–covariance matrix if the variances and correlation coefficients are estimated from all available data on individual variates and pairs of variates respectively. An alternative is to estimate the missing values by regression techniques and to calculate a revised variance–covariance matrix. This method is suitable for use with an electronic computer. It is shown that with this method the resultant covariances are unbiased, but that the variances require correction for bias. A numerical example is given.

Race/Ethnicity and Nonadherence to Prescription Medications Among Seniors: Results of a National Study

“…For approximately 10% of respondents with missing income data, income was imputed based on Buck's Method, which puts respondents into the single most likely category, which has an advantage for ease of understanding over other imputation methods that may classify respondents into multiple categories based on a probability distribution. 25 …”

A systematic review identifies a lack of standardization in methods for handling missing variance data

“…Methods that impute study-level SD either assume that the SD is associated with other study-level variables or the treatment or control mean (i.e., linear or Buck's regression [14], coefficient of variation [15]), or they assume simply that the missing SD is similar to other study SD (e.g., substitution, arithmetic mean).…”

The structure of infant cognition at 1 year

“…To maximize the sample size for factor analysis, we imputed missing values for those infants lacking data on one or more measures, employing the iterative least squares regression procedure of SYSTAT 10 (Buck, 1960). Given that there were mean differences between groups on a number of measures, imputation was done separately for preterms and full-terms.…”