Selecting the Right Statistical Test

1. The Core Decision: Understanding Your Variables

The most critical factor in choosing a statistical test is the type of variable you are measuring. In our previous lesson, we touched on this, but let's formalize the definitions.

To place these definitions in an Ayurvedic research context:

• Quantitative (Continuous) Variables: These are numerical measurements that can take any value within a range.

  • Examples: Hemoglobin level (e.g., 12.5 g/dL), blood pressure (e.g., 120/80 mmHg), weight (e.g., 65.7 kg), a Visual Analogue Scale (VAS) score for pain from 0-100.

• Categorical Variables: These represent groups or categories.

  • Nominal: Categories with no inherent order.
    • Examples: Prakruti (Vata, Pitta, Kapha), Gender (Male, Female), Treatment group (Ayurvedic, Control).
  • Ordinal: Categories with a clear, meaningful order.
    • Examples: Disease severity (Mild, Moderate, Severe), Bala (Avara, Madhyama, Pravara), satisfaction rating (Poor, Fair, Good, Excellent).

This distinction is the root of our decision tree. Now let's look at that tree.

2. A Visual Guide for Choosing a Test

Navigating statistical tests can seem complex, but a flowchart can simplify the process immensely. The flowchart below maps out the decisions based on your research question and variable types. For this lesson, we are focusing on the left side of the chart: "Comparison the difference between groups."

As you can see, after establishing that you want to compare groups, the next questions are:

• Is your outcome variable (the one you are measuring) continuous or categorical?
• Are your groups independent (e.g., a drug group vs. a placebo group) or paired/dependent (e.g., measurements on the same people before and after an intervention)?
• How many groups are you comparing?

Let's use this framework to explore the most common tests.

3. Key Tests for Comparing Groups

This video provides an excellent overview of the entire selection process. We will refer to it as we discuss the specific tests.

A. Comparing a Continuous Outcome (e.g., BMI, Blood Sugar, Pain Score)

This is a very common scenario in clinical research. For example, you want to see if an intervention changes a measurable physiological parameter.

The TileStats video you started watching covers these tests clearly.

• t-tests are for comparing the means of two groups.
• ANOVA (Analysis of Variance) is for comparing the means of three or more groups.
• Non-parametric tests (Mann-Whitney, Wilcoxon, Kruskal-Wallis) are used when your data is not normally distributed or when you have ordinal data. They compare medians or ranks instead of means.

B. Comparing a Categorical Outcome (e.g., Improved: Yes/No)

What if your outcome isn't a measurement but a count? For example, you want to compare the number of patients who reported symptom relief in a treatment group versus a control group.

For this, we use the Chi-Square (${\chi }^2$) Test.

The chi-square test is used to determine if there is a significant association between two categorical variables. It compares the observed frequencies in your sample to the frequencies that would be expected if there were no relationship between the variables.

The following video provides an excellent clinical example of when to choose the chi-square test.

When to use the Chi-Square test:

• Your independent variable is categorical (e.g., Treatment Group vs. Control Group).
• Your outcome variable is also categorical (e.g., Responded vs. Did Not Respond; Prakruti type).

If your sample size is very small, a variation called Fisher's Exact Test is used instead of the chi-square test.

4. Case Study: Statistical Choices in an Ayurvedic Clinical Trial

Theory is essential, but seeing it applied in practice is what solidifies understanding. Let's analyze a real Ayurvedic clinical trial. This study compares an Ayurvedic formulation (Abhraloha) with a standard modern treatment (Ferrous Ascorbate) for Iron Deficiency Anemia (Pandu).

Let's dissect their choices based on what we've learned:

1. Question: How did they compare the final hemoglobin (Hb) levels between the Abhraloha group and the Ferrous Ascorbate group?

   • Outcome Variable: Hb level (continuous).
   • Groups: Two independent groups.
   • Appropriate Test: Independent t-test or Mann-Whitney U test.
   • What they did: They used the Mann-Whitney U test. This tells us their data likely wasn't normally distributed, so they chose the correct non-parametric option.

2. Question: How did they assess if Hb levels significantly increased from baseline to the final visit within the Abhraloha group?

   • Outcome Variable: Hb level (continuous).
   • Groups: Two paired measurements (baseline vs. final) on one group.
   • Appropriate Test: Paired t-test or Wilcoxon Signed-Rank test.
   • What they did: They used the Wilcoxon signed-rank test for within-group analysis.

3. Question: How did they compare the number of patients reporting subjective improvement (a categorical variable) between the two groups?

   • Outcome Variable: Improvement Yes/No (categorical).
   • Groups: Two independent groups.
   • Appropriate Test: Chi-square test or Fisher's exact test.
   • What they did: They used the chi-square test, and noted that Fisher's exact test would be used if cell counts were too low.

This study perfectly illustrates the decision-making process in action.

Conclusion

In this lesson, we have constructed a practical framework for selecting the correct basic statistical test for comparing groups. We learned that this critical decision is not arbitrary but is guided by a logical assessment of your research design and data.

Key Takeaways:

• The choice of a statistical test depends primarily on the type of your outcome variable (continuous vs. categorical), the number of groups being compared, and whether the groups are independent or paired.
• For comparing continuous outcomes between two groups, use a t-test. If the groups are independent, use an independent samples t-test; if they are paired (e.g., before/after), use a paired t-test.
• For comparing continuous outcomes among three or more independent groups, use One-Way ANOVA.
• For comparing categorical outcomes (i.e., frequencies or proportions), use the Chi-Square (${\chi }^2$) test.
• Every parametric test (t-test, ANOVA) has a non-parametric counterpart (Mann-Whitney U, Wilcoxon, Kruskal-Wallis) to be used when the data does not meet certain assumptions like normality.

Preview of the Next Lesson:

Now that you can select the right tool for the job, the next logical step is to understand the output it produces. In our next lesson, we will focus on the learning outcome: "Interpret p-values and confidence intervals from research papers to draw conclusions about study hypotheses." We will delve into what these crucial numbers mean and how to use them to decide whether your Swasthawritta intervention had a statistically significant effect.