Evaluation of statistical methods applied in theses and dissertations in an Open, Distance and e-Learning University

TL;DR

This study evaluates the use of statistical methods in master's and doctoral theses at an African e-learning university, finding many common errors that affect research quality.

Contribution

The study identifies specific statistical method errors in agricultural and environmental science theses, offering insights to improve postgraduate research training.

Findings

41.0% of theses and dissertations contained at least one major statistical or methodological error.

Analysis of variance was the most commonly used statistical test, followed by t-tests and chi-square tests.

Common errors included inappropriate sampling techniques and incorrect statistical modeling.

Abstract

The appropriate application of research methods and statistical analyses used in the studies directly affects the quality of scientific studies. Due to the possibility of employing an incorrect statistical technique, it is crucial to choose a statistical method based on the study’s data and research objectives. This study aimed to evaluate whether statistical techniques applied in the theses and dissertations were appropriate for planning surveys or experiments and analyzing data, and to identify common mistakes that master’s and doctoral students made when using statistical techniques for the intended goals. The study reviewed 139 master’s theses and doctoral dissertations submitted to seven agricultural and environmental sciences disciplines at a leading Open, Distance and e-learning university in Africa between 2015 and 2020. These dissertations and theses used mixed and quantitative research methods. The analysis of variance test was the most often used statistical test, according to the results, followed by the student t-test and the Chi-square test. At least one blatant methodological error was found in 41.0% of theses and dissertations, either in the data collection process or in the data analysis. Examples of these errors include the use of a simple random sampling technique despite the heterogeneous population units, the conversion of count responses to binary responses and percentages for fitting logistic and general binomial regression models, and the incorrect modeling of correlated data using generalized linear models. The results of this study will create greater awareness of the common errors that postgraduate students make when using statistical methods to design experiments or sample surveys and analyse data. In addition, the findings inform the university management to plan for specific training in statistical methods appropriate for a range of academic fields.

Figures