Keywords

Paired t test, Paired data, Unpaired data, Type I error, Equality of two means, Simulation study

Abstract

Practitioners often resort to the paired t test to test equality of two means when data is complete. When small amounts of data are missing, is this method still as robust? Are there alternative methods which can perform the job better on the missing data? In this thesis, we measured the robustness of the paired t-test, two sample t-test, corrected z-test, modified corrected z-test, weighted t-test, pooled t-test, optimal pooled t-test, unequal variance test, correlation focused t-test, multiple imputation method, mixed model method, modified maximum likelihood estimate method, and Uddin’s modified maximum likelihood estimate method. This is simulated in an environment which considers all possible combinations of specified missing values. The results suggest that when the paired t test hovers above the significance level of 0.05 on complete data, some alternative methods perform just as well if not better depending on case-by-case scenarios. For smaller sample sizes, the mixed model method performs the best under a series of strict significance levels. As sample size increases, all methods perform similarly well.

Completion Date

2025

Semester

Spring

Committee Chair

Uddin, Nizam

Identifier

DP0029371

Document Type

Dissertation/Thesis

Download

Share

COinS