Applied Missing Data Analysis (Methodology in the Social Sciences)
by Craig K. Enders
Where could I find the missing data? Applied missing data analysis by Enders, 2010,
I recently dived into problems of missing data. Intuitively, data analysts would be suspicious about missing data and many times would prefer to delete these records as whole, e.g. respondents to longitudinal surveys respond to the first two repeats but not to the third. The questions is what is better, to run more robust likelihood functions (full information maximum likelihood) which tend not to converge, or to impute missing data based on the data themselves? Enders takes us along the process of the later option – how to identify missing data and why to consider replacement, what are the better techniques to impute missing values for each case we handle? Enders supports his text with many examples and charts that explain the process outcome. He also tests the different suggested procedures by means of simulations and shows the power of each and mainly, how good imputation strategy almost does not bias the original data. On the trade-off between larger sample size and small bias, randomly imputing missing data makes a great difference. Statistical models run the variance/covariance matrix, thus the effect of each record is small, that is why imputations adds just little bias to the final model we use. I use the book heavily in my statistical work and find it helpful every time I face problems of missing data and want to convince the collector of the data that imputations are harmless. Dr. Gabriel Liberman –Data-Graph Statistical Consulting at: www.data-graph.com .
ד"ר גבי ליברמן – דטה גרף, מחקר וייעוץ סטטיסטי
אין תגובות