Parametric vs Non-Parametric Statistics
1. Parametric Statistics
Parametric methods assume that the data come from a population that follows a
known distribution, most commonly the normal distribution
• They rely on parameters like mean (µ), variance (σ²), etc. We estimate these
parameters and then make inferences about the population.
• They are generally considered more powerful (i.e., more precise in detecting
true effects) if their assumptions are satisfied.
• However, if the assumptions are violated, the results may be misleading or
incorrect.
Examples: t-test, Z-test, ANOVA, Pearson correlation.
2. Non-Parametric Statistics
Non-parametric methods do not assume that the data follow a specific
distribution. They are often called “distribution-free” tests because they make
fewer assumptions.
• They are especially useful when data are ordinal or nominal, or when the
assumptions of parametric tests (e.g., normality, homogeneity of variance)
are not satisfied.
• They are also helpful with small sample sizes or when there are outliers,
since these can distort parametric methods.
• Non-parametric methods are generally less powerful than parametric ones,
but they are safer and more robust when assumptions are not met.
Examples: Mann-Whitney U test, Wilcoxon signed-rank test, Kruskal-Wallis test,
Chi-square test, Spearman’s rank correlation.
Key Differences Between Parametric and Non-Parametric Tests
Aspect Parametric Non-Parametric
Assumptions Assumes known distribution (often Few assumptions; does not
normal), equal variances, require normality; robust
independent observations. to outliers.
Type of Data Interval/ratio scale (continuous Works with ordinal,
measurements). nominal, ranked data,
small samples.
Statistical More powerful if assumptions are Less powerful, but more
Power satisfied (detects smaller reliable when assumptions
differences). are violated.
Examples t-test, ANOVA, correlation, Mann-Whitney, Wilcoxon,
regression. Kruskal-Wallis, Spearman,
Chi-square.
How to Decide Which Test to Use (Step-by-Step)
1. Check Normality: Use a histogram, Q-Q plot, or formal tests like Shapiro-
Wilk.
2. Check Sample Size: Small samples often make parametric assumptions
unreliable.
3. Check Outliers: Strong outliers can distort mean and variance.
4. Check Measurement Level: If data are ordinal or nominal, parametric
methods may not apply.
5. Check Assumptions: Ensure independence and homogeneity of variance
when required.
➡ If assumptions hold → Use parametric (more powerful).
➡ If assumptions are violated → Use non-parametric (safer).
Hypothesis Testing
Key Terms
Term Meaning Why It’s Important /
Example
Hypothesis (in A statement or assumption about a Example: “The average
statistics) population parameter that we test height of students in my
using data. college is 160 cm.”
Null Hypothesis The default assumption: no effect, Example: H₀: “Mean
(H₀) no difference, no relationship. height = 160 cm.”
Alternative The competing assumption: there is Example: H₁: “Mean
Hypothesis (H₁ / an effect or a difference. height ≠ 160 cm.”
Hₐ)
Level of Probability threshold for rejecting Example: α = 0.05 → 5%
Significance (α) H₀. Common values: 0.05, 0.01, risk of wrongly rejecting
0.10. H₀.
Degrees of The number of values “free” to vary Example: Single-sample
Freedom (df) given constraints (depends on t-test → df = n – 1.
sample size & parameters
estimated).
Hypothesis Testing Flow
1. State H₀ and H₁.
2. Choose α (e.g., 0.05).
3. Collect data and compute test statistic (t, Z, χ², etc.).
4. Find p-value or critical value.
5. Decision:
o If p ≤ α → Reject H₀ (evidence supports H₁).
o If p > α → Fail to reject H₀ (not enough evidence).
6. Conclude in context and consider possible Type I or II errors.
Common Mistakes in Hypothesis Testing
• You never “prove” H₀ — you only fail to reject it.
• Don’t choose α after seeing the data.
Miscalculating degrees of freedom leads to wrong results
• Lower α = stricter test → fewer Type I errors, but more risk of Type II errors.
One-Tailed vs Two-Tailed Tests
Two-Tailed Test (દ્વિ-પછ
ૂં ડી કસોટી)
• Used when you only want to know if there is a difference, not its direction.
• The rejection region is split between both tails of the distribution.
• Example: Testing whether average height ≠ 160 cm.
One-Tailed Test (એક-પછ
ૂં ડી કસોટી)
• Used when you predict a specific direction of difference.
• The rejection region lies entirely in one tail.
• Example: Testing whether average height > 160 cm.
ANOVA & Experimental Designs
Two-Way ANOVA (દ્વિમાર્ગીય દ્વિચરણ દ્વિશ્લેષણ)
• Tests the effect of two independent variables on one dependent variable.
• Examines:
1. The main effect of each independent variable.
2. The interaction effect (when the effect of one variable depends on the
level of the other).
Factorial Design (ઘટક યોજન)
• An experimental design where all possible combinations of levels of
independent variables are tested.
• Example: A 2 × 2 design has 2 independent variables, each with 2 levels.
• Advantages: Efficient and helps detect interactions between variables.
Extra Definitions
• Rejection Region: The range of values where H₀ is rejected.
• Distribution: A function showing possible values & frequencies (e.g., normal
distribution = bell curve).
• Independent Variable (IV): The variable that is manipulated or controlled.
• Dependent Variable (DV): The outcome measured; depends on the IV.
• Interaction Effect: When the impact of one IV depends on the level of
another IV.