A key part of statistical analysis and hypothesis testing in biomedical research is to explore relationships between variables and differences between groups. Such statistical tests are grouped under the term “inferential statistics.”
What are inferential statistics?
Inferential statistics are a way of drawing inferences (i.e., conclusions) about a population based on a smaller subset, known as a sample. As mentioned above, these conclusions or inferences are typically about whether certain variables are related to each other or whether certain groups differ from each other. Inferential statistics enable us to determine whether these variables or relationships are due to chance (i.e., whether they are statistically significant).
Differences between Descriptive and Inferential Statistics
Descriptive statistics are different from inferential statistics in that the former is mainly used to summarize the data (measures of central tendency and variance). They give us a consolidated picture of the data, but we usually can’t draw conclusions from descriptive statistics alone.
Descriptive statistics enable us to organize and present data in a meaningful manner so that we can run inferential statistical analyses. Inferential statistics enable us to make comparisons, hypotheses, and predictions.
Finally, descriptive statistics focus on what is known (i.e., data already collected). Inferential statistics allow us to infer about the population from which our sample was drawn, without having to collect data for the entire population.
Types of Inferential Statistics
There are numerous types of Inferential analyses available. They are divided into two main groups:
- Testing for relationships: E.g., correlation analysis, regression analysis
- Testing for differences: E.g., ANOVA, Wilcoxon signed-rank test, Student’s t-test
Choosing the right inferential statistics
Since inferential statistics are a means of testing relationships or differences among your study variables, their output is often the most important result of your study. Hence, it’s particularly important to choose your inferential statistics carefully. Here are some tips to help you choose the right inferential statistical test:
- Refer to your research question and study aims: Based on what you are investigating, you can choose whether you want to compare two groups or test the relationship between two variables
- Check how many variables you are analyzing at a time: Certain tests, such as the t-test, can be done with only two variables. Others like ANOVA or MANOVA can be used with multiple variables.
- Check the type of data: Whether the variables you are analyzing are categorical, continuous, or nominal will also affect the choice of test. For example, the relationship between two continuous variables can be tested using Pearson’s correlation analysis or Spearman’s correlation analysis, but if the variables are categorical, you will need to perform a chi-square test of association.
- Check the distribution of continuous data: If your data is normally distributed, you can run tests like Student’s t-test, but if the data is nonparametric, you will need to run a Mann-Whitney U test or a Kruskal-Wallis test.
- When comparing groups, check if data are paired: If the two variables are independent of each other (i.e., come from different samples), you can use a test like the unpaired t-test, but if they are paired (i.e., come from the same samples at different points in time), you need to run a paired t-test.
- Inspect your data for outliers: Outliers can affect some of your descriptive and inferential statistics. Outliers that are clearly due to errors in measurement or data entry (e.g., platelet count of 5) can be removed, though you will need to report having done so in your research paper.
Would you like expert guidance on selecting the right inferential statistics to strengthen your statistical analysis? Editage’s Statistical Analysis & Review Services can help!
Comment