How do I do dissertation data analysis?
Data Analysis Plan Overview
Dissertation methodologies require a data analysis plan.Your dissertation data analysis plan should clearly state the statistical tests and assumptions required to examine each research question, explain how you clean and create scores, and specify the desired sample size for each test. You should select statistical tests based on two factors: (1) how you phrase the research questions and hypotheses, and (2) the level of measurement of the variables. For example, if you examine the impact of variable X on variable Y, you would use regressions. If you seek associations or relationships, you would use correlation and chi-square tests. If you examine differences, t-tests and ANOVAs are likely the correct tests.
Level of Measurement
The level of measurement is the second factor in selecting the correct statistical test. If the research question examines the impact of X on Y, and Y is scale, use linear regression (e.g., the impact of income on savings) .If Y is ordinal, you should use ordinal regression. Conversely, if both X and Y are categorical, chi-square is the appropriate test. Ultimately, both the phrasing of the research question and the level of measurement guide the test selection.
Statistical Assumptions in Data Analysis Plan
Part of the data analysis plan is to document the assumptions of a particular statistical test. Most assumptions fall into the normality, homogeneity of variance, and outlier bucket of assumptions. Other tests have additional assumptions. For example, in a linear regression with several predictors, the variance inflation factor needs to be assessed to determine that the predictors are not too highly correlated. This data analysis plan video may be helpful.
Composite Scores and Data Cleaning
Data analysis plans should discuss any reverse coding of the variables and the creation of composite or subscale scores. You should plan to examine alpha reliability before creating composite scores. Additionally, document the data cleaning procedure. For example, the removal of outliers, transforming variables to meet normality assumption, etc.
Sample Size and Power Analysis
After selecting the appropriate statistical tests, data analysis plans should follow-up with a power analysis. The power analysis determines the sample size for a statistical test, given an alpha of .05, a given effect size (small, medium, or large) at a power of .80 (that is, an 80% chance of detecting differences or relationships if in fact difference are present in the data. This power analysis video may be helpful.