Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


School of Education


The present study deals with the problem of comparison between a two medical facilities' with extremely skewed sample sizes from non-experimental study. The data came from a study of rehabilitation interventions with patients diagnosed with cardiac and pulmonary issues who received treatment either in inpatient rehabilitation facilities (IRFs) or in skilled nursing facilities (SNFs). Due to inclusion and exclusion criteria, however, the study had failed to recruit sufficient number of participants between two comparison groups: 319 from IRFs and 27 from SNFs. As a result, the main hypothesis of the study was not tested due to the disparity of the participants between the two comparison groups, which could not be analyzed as a study with an unbalanced design because of lack of power in the analysis (Beacham, 2008).

In medical research, this kind of problem occurs often not only because of inclusion and exclusion criteria in recruiting patients for a study but also because of dropout patients due to many reasons, such as technical changes (certain insurance and/or Medicare policies eliminate possible participants), medical changes, or personal circumstances change in the middle of the study. By extracting matching methods from both Fisher's experimental design and Rubin's Causal Model (RCM) the present study attempts to offer ways to draw the causal inference in a non-experimental study with sample size disparity between two comparison groups, especially when collected data disable a researcher to analyze.

The matched datasets were analyzed in two ways: multivariate of covariance (MANCOVA) first and two analysis of covariance (ANCOVA) models when there was a significant main effect in the previous MANCOVA model. No significant different effectiveness was found between IRFs and SNFs in the 1:1 Matched Data, but IRFs took better care than SNFs in the Caliper Matched Data, rehabilitating the patients diagnosed with cardiac and pulmonary diseases on the functional independent measure (FIM). In comparison methodology, the results suggested that datasets created by both matching methods provided similar results, but that Fisher's design fits better for small dataset while RCM, for larger dataset by using propensity scores to balance the matching sets.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.