Selecting statistical tests for your quantitative research

Selecting a statistic test

In the measures, we define and identify our outcomes or endpoints. We can identify one outcome or endpoint as our primary outcome and we can have several other secondary outcomes or endpoints

We will look at tests pertaining to quantitative research and focus on some of the basic or commonly used tests. We will not run a discussion of the theoretical background of statistical tests in this module.

You can use this module to identify some of the common tests that you can employ in your research (and based on your research design and question)

Note:  Earlier, we mentioned the need to note down anticipated answers. Your research question and anticipated answers will help you plan the tests that will be useful for your research. Please be aware that your anticipated choice of test may change after you conduct your research based on the data you obtain.

 

STEP 1: IDENTIFY THE DEPENDENT VARIABLE OF INTEREST

STEP 2: IDENTIFY THE INDEPENDENT VARIABLE(S) OF INTEREST

Identify One Dependent Variable
No Independent Variable Univariate Analysis Can be

  1. Continuous Dependent Variable
  2. Ordinal Dependent  Variable
  3. Nominal Dependent Variable
One Independent Variable Bivariate Analysis Can be

  1. Continuous Dependent Variable
  2. Ordinal Dependent  Variable
  3. Nominal Dependent Variable
More than one Independent Variable Multivariate Analysis Can be

  1. Continuous Dependent Variable
  2. Ordinal Dependent  Variable
  3. Nominal Dependent Variable

 

Let us look at each of these scenarios

Univariate Analysis

Assumptions

  1. Population from which sample is drawn is normally distributed
  2. Sample (s) are randomly selected from the population
  3. Samples have approximately equal variance (homogeneity of variance)
  4. Parametric data- based on an interval or ratio based scale
Continuous Dependent Variable & No Independent Variable Comparing Means Students t-test
Continuous Dependent Variable & No independent Variable comparing one pre-and one post measure in same subject Paired t test

 

Assumptions

  1. Distribution free or non-parametric data
  2. Samples are randomly selected or representative of a general population
Ordinal Dependent Variable & No independent Variable Comparing Medians Wilcoxon Sign Rank test
Nominal Dependent Variable Affected by Time & no independent variable Rate Normal Approximation to Poisson
Nominal Dependent Variable NOT affected by timeOutcome Common & No independent Variable Proportion Normal Approximation to Binomial
Nominal Dependent Variable NOT affected by timeOutcome Uncommon &  No independent Variable Proportion Normal Approximation to Poisson

 

Bivariate Analysis

Ordinal Dependent Variable & No independent Variable Comparing Medians Wilcoxon Sign Rank test
Nominal Dependent Variable Affected by Time & no independent variable Rate Normal Approximation to Poisson
Nominal Dependent Variable NOT affected by time- Outcome Common & No independent Variable Proportion Normal Approximation to Binomial
Nominal Dependent Variable NOT affected by time- Outcome Uncommon &  No independent Variable Proportion Normal Approximation to Poisson
Ordinal Dependent Variable & No independent Variable Comparing Medians Wilcoxon Sign Rank test
Nominal Dependent Variable Affected by Time & no independent variable Rate Normal Approximation to Poisson
Nominal Dependent Variable NOT affected by time- Outcome Common & No independent Variable Proportion Normal Approximation to Binomial
Nominal Dependent Variable NOT affected by time- Outcome Uncommon &  No independent Variable Proportion Normal Approximation to Poisson
Continuous Dependent Variable & Continuous Independent variable from naturalistic or purposive sampling Slope and Intercept Regression Analysis, F-Test
Continuous Dependent Variable & Continuous Independent Variable from naturalistic sample Pearson’s correlation coefficient Correlation analysis, Students t-test
Continuous Dependent Variable & Nominal Independent Variable from naturalistic sample or purposive sample Difference between means Students t-test
Ordinal Dependent Variable & Ordinal Independent Variable from naturalistic sample Spearmans Correlation Coefficient Spearmans Correlation test
Ordinal Dependent Variable & Nominal Independent Variable from a naturalistic or purposive sample Difference between medians Mann Whitney Test
Nominal dependent Variable NOT AFFECTED by time  & Nominal Independent Variable- paired design Odds Ratio, Relative Risk, difference in proportions McNemars test
Nominal dependent Variable NOT AFFECTED by time & Nominal Independent Variable- unpaired design Odds Ratio, Relative Risk, difference in proportions Chi-square test, normal approximation, Mantel Haenzel Test, Fishers Exact Test
Nominal dependent Variable AFFECTED by time & Nominal Independent Variable- unpaired design Rate difference or ratio Normal approximation
Nominal dependent Variable NOT AFFECTED by time & Continuous Independent Variable Slope and intercept Chi-square test for trend

 

Multivariate Analysis

Continuous Dependent Variable & Nominal Independent variables representing one characteristic Means One-Way Analysis of Variance (ANOVA), F test, Student Newman Keuls Test
Continuous Dependent Variable & Nominal Independent variables representing > 1 characteristic Means Factorial Analysis of Variance (ANOVA), F test, Student Newman Keuls Test
Continuous Dependent Variable & Continuous Independent Variables from naturalistic sample or purposive sample Regression coefficients Multiple regression analysis, F test, partial F test
Continuous Dependent Variable & Continuous Independent Variables from naturalistic sample Coefficient of determination Multiple correlation analysis, F test
Continuous Dependent Variable & Nominal or continuous Independent Variables from a naturalistic or purposive sample Regression coefficients Analysis of Covariance (ANCOVA), F test, Partial F test
Ordinal Dependent Variable & Ordinal Independent Variables Kendalls coefficient of concordance Chi-square test
Ordinal Dependent Variable & Nominal Independent Variables Mean of Ranks Kruksal Wallis Test or Dunn’s test
Nominal dependent Variable AFFECTED by time & Continuous and Nominal Independent Variables Incidence Ratio Cox Proportional Hazards Regression, Chi-square test
Nominal dependent Variable NOT AFFECTED by time & Continuous and Nominal Independent Variables Odds Ratio Logistic Regression, chi square test
Nominal dependent Variable AFFECTED by time & Nominal independent Variables Cumulative probabilities Kaplan Meier Life table analysis, Mantel Haenzel Test or logrank test

Nominal dependent Variable NOT AFFECTED by time & Nominal independent Variables

Odds ratio, relative risk or difference in proportions

Stratified analysis, Mantel Haenzel test

In subsequent posts, we will look at how to do each of these tests and interpret them, using real life data examples.
Post attributed to Research Action Group AMMA ERF