Abstract:
|
Due to massive data collection, missing data are highly likely to be present in data analysis.They could lead to wrong decisions in statistical hypotheses tests and results misinterpretation. Additionally, the missing data undermine the power of sample sizes designed to achieve certain effect sizes. In this study we analyze data on Methicillin-resistant Staphylococcus aureus(MRSA) and methicillin susceptible strains(MSSA) recurrence from a collaborative research that investigated a sample from 6 Community Health Centers in the New York area.In a questionnaire about demographics, submitted to 129 subjects, it was observed 11.7% rate of missing-ness when respondents were asked about their immigrant status. Since it is of primary interest correlate this factor with MRSA and MSSA recurrence, a sensitivity analysis was carried out under two different missing-data mechanisms: MCAR(Missing Completely at Random) and MAR(Missing at Random). In MCAR, simply random number generation from Bernoulli distribution was applied for imputation. Under MAR mechanism, tree-based methods were applied for multiple imputation. Stability of the p-values is presented as measure of missing data impact.
|