by Selecting an appropriate statistical technique is a critical decision that determines whether research conclusions are valid or misleading. The choice depends on the research objective (description, comparison, association, prediction), measurement level of variables (nominal, ordinal, interval, ratio), number of groups or variables, and underlying assumptions (normality, homogeneity of variance). In Indian business research, common errors include using parametric tests on non normal data, applying correlation for categorical variables, or choosing overly complex techniques for simple questions. A wrong statistical technique produces incorrect p values, wrong conclusions, and failed replication. Selection must be justified in the methodology section.
1. Research Objective – Description
When the research objective is purely descriptive (summarizing data without testing relationships or differences), use descriptive statistics. For central tendency: mean (interval/ratio data), median (ordinal or skewed data), mode (nominal data). For dispersion: range, standard deviation, variance (interval/ratio); interquartile range (ordinal); frequency distribution (nominal). In Indian business research, describing ecommerce customer age distribution requires mean and standard deviation. Describing preferred payment methods requires frequencies and percentages. No inferential tests are needed. The key principle: match the descriptive statistic to the measurement level. Do not report mean for nominal variables (e.g., average gender). Descriptive statistics are the foundation before any inferential testing. Always report appropriate descriptive statistics for each variable before hypothesis testing.
2. Research Objective – Comparison (Two Groups)
When comparing two independent groups (e.g., men vs women, urban vs rural), use independent samples t test if the dependent variable is interval/ratio and assumptions (normality, homogeneity of variance) are met. If assumptions are violated, use Mann Whitney U test (non parametric). For two paired groups (same respondents before and after, e.g., satisfaction before and after training), use paired samples t test (parametric) or Wilcoxon signed rank test (non parametric). In Indian business research, comparing ecommerce spending between two cities (Delhi vs Mumbai) requires independent t test. Comparing spending before and after Diwali sale requires paired t test. Check assumptions before selecting the test. Reporting wrong test invalidates conclusions. Always state which test was used and why.
3. Research Objective – Comparison (Three or More Groups)
When comparing three or more independent groups (e.g., satisfaction across five ecommerce platforms), use one way ANOVA (parametric) if dependent variable is interval/ratio and assumptions hold. If ANOVA is significant, follow with post hoc tests (Tukey, Bonferroni) to identify which groups differ. If assumptions are violated, use Kruskal Wallis H test (non parametric). For repeated measures (same subjects measured multiple times), use repeated measures ANOVA (parametric) or Friedman test (non parametric). In Indian business research, comparing monthly spending across income groups (low, medium, high) requires one way ANOVA. Do not perform multiple t tests; this inflates Type I error. ANOVA controls overall error rate. Report effect size (eta squared) alongside p value.
4. Research Objective – Association (Two Variables)
When examining the relationship between two variables, select based on measurement level. Both variables interval/ratio: use Pearson correlation coefficient (r). Values range from -1 to +1. Both variables ordinal: use Spearman rank correlation (rho). One nominal, one interval/ratio (categorical vs continuous): use point biserial correlation (binary nominal) or eta (multicategory nominal). Both nominal: use contingency coefficient or Cramer’s V based on chi square. In Indian business research, correlating age (interval) with ecommerce spending (interval) requires Pearson. Correlating satisfaction rank (ordinal) with income rank (ordinal) requires Spearman. Correlation does not imply causation. Report both correlation coefficient and p value. Also report sample size because correlation stability depends on n. Scatterplots should accompany correlation analysis to detect non linear relationships.
5. Research Objective – Prediction (Regression)
When predicting a dependent variable from one or more independent variables, use regression analysis. For one continuous dependent variable and one independent variable: simple linear regression. For one continuous dependent variable and multiple independent variables: multiple linear regression. For binary dependent variable (yes/no, buy/not buy): logistic regression. For ordinal dependent variable: ordinal logistic regression. In Indian business research, predicting ecommerce spending from age, income, and location requires multiple linear regression. Predicting purchase decision (buy vs not buy) from same predictors requires logistic regression. Check assumptions: linearity, independence, homoscedasticity, normality of residuals, no multicollinearity. Report R squared (variance explained), unstandardized coefficients (B), standardized coefficients (beta), p values, and confidence intervals. Regression is for prediction, not just association.
6. Measurement Level of Dependent Variable
The dependent variable’s measurement level is the primary determinant of statistical technique. For interval/ratio dependent variables (e.g., spending in rupees, satisfaction score), use parametric tests: t test, ANOVA, Pearson correlation, linear regression. For ordinal dependent variables (e.g., Likert scale, rank order), use non parametric tests: Mann Whitney, Kruskal Wallis, Spearman correlation, ordinal regression. For nominal dependent variables (e.g., brand choice, purchase yes/no), use chi square, logistic regression, or log linear analysis. In Indian business research, a common mistake is treating Likert scale (ordinal) as interval and using parametric tests without justification. While many researchers do this, assumptions must be checked and defended. If in doubt, use non parametric tests. They are safer but less powerful. Always state measurement level and justify test selection.
7. Number and Type of Independent Variables
The number and nature of independent variables determine which statistical model to use. One independent variable with two groups: t test or Mann Whitney. One independent variable with three or more groups: ANOVA or Kruskal Wallis. One continuous independent variable: correlation or simple regression. Two or more independent variables (all categorical): factorial ANOVA or log linear analysis. Two or more (mixed categorical and continuous): ANCOVA or multiple regression. Two or more (all continuous): multiple regression. In Indian business research, studying effect of training (yes/no) and gender (male/female) on sales requires factorial ANOVA. Studying effect of price (continuous) and advertising (continuous) on sales requires multiple regression. Do not use simple tests for complex designs. Do not use complex tests when simple tests suffice. Match model complexity to research design.
8. Assumptions of Parametric Tests
Parametric tests (t test, ANOVA, Pearson, regression) assume: (1) normality of data or residuals, (2) homogeneity of variance across groups, (3) independence of observations, (4) for regression: linearity and no perfect multicollinearity. Violating assumptions increases Type I or Type II errors. In Indian business research, check normality using Shapiro Wilk test, Q Q plots, or skewness/kurtosis values. For large samples (n > 100), normality violations are less serious due to central limit theorem. Check homogeneity using Levene’s test. If assumptions are violated, use non parametric alternatives (Mann Whitney instead of t test, Kruskal Wallis instead of ANOVA, Spearman instead of Pearson). Do not ignore assumptions. Report assumption checks in methodology. If assumptions are violated but you use parametric tests anyway, justify with central limit theorem or robustness evidence.
9. Sample Size Considerations
Sample size affects statistical power and determines which tests are feasible. For parametric tests (t test, ANOVA, regression), minimum sample size guidelines: t test requires at least 30 per group; ANOVA requires at least 20 per group; regression requires 10 to 20 cases per independent variable. For non parametric tests, larger samples are generally needed to achieve equivalent power. For chi square tests, expected frequency in each cell should be at least 5. In Indian business research, small samples (n < 30) should use exact tests or non parametric methods. Very large samples (n > 1000) detect even trivial differences as statistically significant; report effect sizes alongside p values. Use power analysis (G*Power) before data collection to determine required sample size for your planned tests. Underpowered studies waste resources; overpowered studies find meaningless significance.
10. Parametric vs Non-Paramratic Tests
Parametric tests assume specific population distributions (usually normal) and are more powerful when assumptions hold. Non parametric tests make no distributional assumptions but have lower power (need larger samples to detect same effect). Use parametric when: data are interval/ratio, normally distributed, variances homogeneous, and sample size adequate. Use non parametric when: data are ordinal, severely non normal, small sample (n < 30), or assumptions violated. In Indian business research, Likert scale data are ordinal but often treated as parametric with justification. For small samples or skewed distributions, use non parametric. Examples: Mann Whitney instead of t test, Kruskal Wallis instead of ANOVA, Spearman instead of Pearson, Wilcoxon instead of paired t test. Non parametric tests are safer but less familiar to some readers. Report both parametric and non parametric results if assumptions are borderline.
