HYPOTHESIS TESTING IN

ECONOMETRICS
A Complete Topper's Reference Guide
Based on Wooldridge – Introductory Econometrics (2012)
Covers: Sampling | Distributions | t-test | z-test | Confidence Intervals | F-test | p-values | Two-tailed &
One-tailed Tests

1. CONCEPTUAL FOUNDATIONS

1.1 What is Hypothesis Testing?

Hypothesis testing is a formal statistical procedure that allows us to make decisions about population
parameters using sample data. In econometrics, we never observe the entire population — we draw a
random sample, compute estimates, and then ask: is what we observe in our sample due to a true
population effect, or is it just random chance?
The Core Question: Given that our OLS estimate β̂ⱼ differs from some hypothesised value, is that
difference statistically meaningful or could it plausibly be sampling noise?

1.2 Population Model Recap

The regression model in the population is:
Population Model y = β₀ + β₁x₁ + β₂x₂ + ... + βₖxₖ + u
β₀, β₁, …, βₖ are unknown population parameters. u is the unobserved error. We estimate these
using OLS on a random sample.

OLS Estimator β̂ⱼ is the estimate of the true population parameter βⱼ. It varies across samples (it is a
random variable). Hypothesis testing is about drawing conclusions from this random variable.

1.3 The Gauss-Markov Assumptions (MLR.1–MLR.5)

The validity of hypothesis testing in OLS rests on five assumptions:
Assumption Name What It Says
MLR.1 Linearity in y = β₀ + β₁x₁ + ... + βₖxₖ + u in the population
Parameters
MLR.2 Random Sampling We have n observations drawn randomly from the population
MLR.3 No Perfect No explanatory variable is a perfect linear combination of
Collinearity others
MLR.4 Zero Conditional E(u|x₁, …, xₖ) = 0 — the error is unrelated to the regressors
Mean
 MLR.5 Homoskedasticity Var(u|x₁, …, xₖ) = σ² — constant error variance
MLR.6 Normality of Errors u ~ Normal(0, σ²) — needed for exact finite-sample inference

Key insight: MLR.1–MLR.4 give unbiasedness of OLS. Adding MLR.5 gives the variance formula and
BLUE (Best Linear Unbiased Estimator). Adding MLR.6 gives exact t and F distributions needed for
hypothesis testing.
2. SAMPLING & DISTRIBUTIONS

2.1 Sampling Distribution of β̂

Because we sample randomly, our OLS estimate β̂ⱼ is a random variable — it would take a different
value in each repeated sample. The collection of all these values across all possible samples is the
Under MLR.1–MLR.6 (Classical Linear Model assumptions):
Key Result β̂ⱼ ~ Normal(βⱼ, Var(β̂ⱼ))
The OLS estimator is normally distributed, centred on the true value βⱼ, with variance given by the
formula below.

Variance of β̂ⱼ Var(β̂ⱼ) = σ² / [SSTⱼ(1 − Rⱼ²)]

SSTⱼ = total sample variation in xⱼ; Rⱼ² = R-squared from regressing xⱼ on all other regressors; σ² =
error variance

What affects the size of Var(β̂ⱼ)?

Factor Effect on Variance Practical Implication
↑ σ² (more noise) ↑ Variance More noise in the model → harder to pin down βⱼ
precisely
↑ SSTⱼ (more ↓ Variance More variation in the regressor → more information
variation in xⱼ) → smaller variance. This is why larger samples
help.
↑ Rⱼ² (high ↑ Variance When xⱼ is highly correlated with other regressors,
multicollinearity) Rⱼ² → 1, variance → ∞

2.2 The Standard Error of the Estimator

Since σ² is unknown in practice, we estimate it with the sample variance of residuals:
Estimated Error σ̂² = SSR / (n − k − 1)
Variance
SSR = Sum of Squared Residuals; n = sample size; k = number of regressors (excluding intercept).
The denominator (n−k−1) are the degrees of freedom.

Standard Error of β̂ⱼ se(β̂ⱼ) = √[ σ̂² / (SSTⱼ(1 − Rⱼ²)) ]

This is the estimated standard deviation of β̂ⱼ. It measures precision of the estimate. Smaller se →
more precise estimate.
2.3 The Normal (z) Distribution — When σ² is Known
When the population error variance σ² is
Standardisation z = (X̄ − μ) / (σ/√n) ~ N(0,1)
X̄ is the sample mean, μ is the population mean under H₀, σ is the known population standard
deviation.

The standard normal N(0,1) has:

• Mean = 0, Variance = 1
• Symmetric bell shape
• P(−1.96 < z < 1.96) = 0.95 (most important critical value)
• P(−2.576 < z < 2.576) = 0.99

Key Feature: Using z vs t

Use z: When σ² is KNOWN (rare) or sample is very large (n > 120) by CLT
Use t: When σ² is UNKNOWN (the normal case in regression)
As n → ∞, the t distribution approaches the z distribution (they are essentially the same for large
samples).

2.4 The t Distribution — When σ² is Unknown

In virtually all regression work, σ² is unknown and must be estimated. This adds additional uncertainty,
captured by the t distribution.
Standardised Estimator t = (β̂ⱼ − βⱼ) / se(β̂ⱼ) ~ t(n−k−1)
(t-statistic)
Under MLR.1–MLR.6 and H₀: βⱼ = value, this follows a t distribution with (n−k−1) degrees of freedom.

Features of the t distribution:

• Bell-shaped and symmetric around zero
• Heavier tails than the normal — reflects extra uncertainty from estimating σ²
• Parameterised by degrees of freedom df = n − k − 1
• As df → ∞, t → N(0,1)
• For df ≥ 30, very similar to normal; for small df (< 10), tails are much fatter

Degrees of Freedom 5% Critical Value 1% Critical Value Note

(two-tailed) (two-tailed)
5 2.571 4.032 Very fat tails
10 2.228 3.169 Still notably fat
20 2.086 2.845 Approaching normal
40 2.021 2.704 Close to normal
120 1.980 2.617 Very close to z
∞ (normal) 1.960 2.576 Standard normal
3. THE HYPOTHESIS TESTING FRAMEWORK

3.1 Types of Hypotheses

Every hypothesis test involves two competing hypotheses:
Null Hypothesis (H₀) Alternative Hypothesis (H₁ or Hₐ)

Definition The baseline claim we test against. What we conclude if evidence against
Usually a restrictive statement (e.g. βⱼ = H₀ is strong enough
0 meaning no effect)
Decision We either REJECT or FAIL TO We reject in favour of H₁ when the test
REJECT H₀ (we never 'accept' H₀) statistic falls in the critical region

Types of Alternative Hypotheses

Type Notation Use When
Two-tailed (two-sided) H₁: βⱼ ≠ 0 You have no prior theory about the sign of βⱼ.
Most common in applied work — reject if |t| >
critical value.
One-tailed (right) H₁: βⱼ > 0 Theory predicts a positive effect. Reject only if t
> critical value (right tail). More powerful if you
are right.
One-tailed (left) H₁: βⱼ < 0 Theory predicts a negative effect. Reject only if
t < −critical value (left tail).

3.2 The Step-by-Step Process of a Hypothesis Test

6 Steps for Any Hypothesis Test
1. STATE H₀ and H₁: Clearly write the null and alternative hypotheses
2. CHOOSE SIGNIFICANCE LEVEL α: Typically 5% (0.05), sometimes 1% or 10%
3. COMPUTE TEST STATISTIC: t = (β̂ⱼ − β⁰) / se(β̂ⱼ) where β⁰ is the value in H₀
4. FIND CRITICAL VALUE: From t-table with df = n − k − 1 at chosen α
5. DECISION RULE: Reject H₀ if |t| > critical value (two-tailed) or if t > c / t < −c
(one-tailed)
6. INTERPRET: Say whether the variable is statistically significant and what it
means economically

3.3 The t-Test for a Single Coefficient

This is the most common hypothesis test in econometrics. We test whether a single explanatory
variable has a statistically significant effect on y.

Most Common Null Hypothesis: H₀: βⱼ = 0 (the variable xⱼ has no partial effect on y)
 t-Statistic t = β̂ⱼ / se(β̂ⱼ)
When H₀: βⱼ = 0. This is what software always reports. Under H₀ and MLR.1–MLR.6, t ~ t(n−k−1).

General form: For H₀: βⱼ = aⱼ where aⱼ is any hypothesised value:

General t-Statistic t = (β̂ⱼ − aⱼ) / se(β̂ⱼ)
Example: Testing H₀: β₂ = 1 (e.g. constant returns to scale) uses t = (β̂₂ − 1) / se(β̂₂)

Critical Regions and Rejection Rules

Test Type Reject H₀ When: Critical Value at 5%
Two-tailed H₁: βⱼ ≠ 0 |t| > c_{α/2} c = 2.00 (large n); check t-table for exact df.
Both tails have area α/2 = 2.5%.
Right-tailed H₁: βⱼ > 0 t > c_α c = 1.645 (large n). Only the right tail has area
α = 5%. LESS conservative than two-tailed.
Left-tailed H₁: βⱼ < 0 t < −c_α c = −1.645 (large n). Only the left tail has area
α = 5%.

Interpreting Decisions Correctly

Terminology — Get This Right!

REJECT H₀ → "The variable is statistically significant at the α% level"
FAIL TO REJECT H₀ → "The variable is statistically insignificant" (NOT "we accept H₀"
— absence of evidence ≠ evidence of absence)
IMPORTANT: Statistical significance ≠ Economic importance. A variable can be
highly significant but have a tiny practical effect. Always check the magnitude of β̂ⱼ.
4. TESTING SIGNIFICANCE OF A COEFFICIENT

4.1 Testing for Statistical Significance (H₀: βⱼ = 0)

The most common test in regression output. Nearly all econometrics software prints this automatically.

Complete Worked Example: Wage Equation

Model: log(wage) = β₀ + β₁educ + β₂exper + β₃tenure + u
Sample results: β̂₁ = 0.092, se(β̂₁) = 0.007, n = 526, k = 3
Step 1: H₀: β₁ = 0 vs H₁: β₁ ≠ 0 (two-tailed)
Step 2: α = 0.05
Step 3: t = 0.092 / 0.007 = 13.14
Step 4: df = 526 − 3 − 1 = 522 → critical value ≈ 1.96
Step 5: |13.14| > 1.96 → REJECT H₀
Step 6: Education is highly statistically significant. Each additional year of education is
associated with approximately 9.2% higher wages, holding experience and tenure fixed.

4.2 One-Tailed Tests: When Theory Guides Direction

Use a one-tailed test when economic theory predicts the

Example: Testing a Positive Effect of Education on Wages

H₀: β₁ ≤ 0 vs H₁: β₁ > 0 (we expect education raises wages)
Critical value at α = 5%, one-tailed, large df → c = 1.645
t = 13.14 > 1.645 → REJECT H₀ in favour of β₁ > 0
Note: The one-tailed critical value (1.645) is smaller than the two-tailed (1.96), so it is easier to reject
when theory is on your side.

Caution on One-Tailed Tests

Only use a one-tailed test if your theory strongly predicts a specific direction BEFORE
seeing the data. Using one-tailed tests to chase significance is p-hacking — a form of data
manipulation.
If the coefficient goes in the wrong direction and you use a one-tailed test, you cannot reject
H₀ even if the t-statistic is very large (in the wrong tail).

4.3 p-Values: The Modern Approach

Instead of comparing a statistic to a fixed critical value, the p-value answers:
"Given H₀ is true, what is the probability of observing a t-statistic as extreme as (or more extreme than)
what we actually got?"
 p-value (two-tailed) p-value = P(|T| > |t|) = 2 × P(T > |t|)
Where T ~ t(n−k−1). A small p-value means the observed data is unlikely under H₀.

Decision Rule: Reject H₀ if p-value < α

p-value range Decision at α=5% Interpretation
p < 0.01 Reject H₀ *** Very strong evidence against H₀. Highly
statistically significant.
0.01 ≤ p < 0.05 Reject H₀ ** Strong evidence against H₀. Statistically
significant at 5%.
0.05 ≤ p < 0.10 Fail to reject Marginal evidence. Insignificant at 5%, but
(borderline) significant at 10%.
p ≥ 0.10 Fail to reject H₀ Little to no evidence against H₀. Variable is
insignificant.

Advantage of p-values: They communicate the strength of evidence continuously, rather than just a
binary reject/fail-to-reject. Preferred in published research.
5. CONFIDENCE INTERVALS

5.1 What is a Confidence Interval?

A confidence interval provides a range of plausible values for the true population parameter β ⱼ, based
on our sample estimate. It is an alternative to the hypothesis test for drawing inferences.

Confidence Interval for β̂ⱼ − t_{α/2} × se(β̂ⱼ) ≤ βⱼ ≤ β̂ⱼ +

βⱼ t_{α/2} × se(β̂ⱼ)
t_{α/2} is the critical value from the t(n−k−1) distribution with α/2 in each tail. For 95% CI at large n:
t_{0.025} ≈ 1.96.

For a 95% confidence interval: Use t_{0.025,df}. The interval is:

95% CI [ β̂ⱼ − 1.96 × se(β̂ⱼ), β̂ⱼ + 1.96 ×
se(β̂ⱼ) ]
(approximately; use exact t critical value for small samples)

5.2 Interpreting a Confidence Interval

Correct vs. Incorrect Interpretation
✓ CORRECT: "If we were to repeat the sampling procedure many times and construct a
95% CI each time, 95% of those intervals would contain the true parameter βⱼ."
✗ WRONG: "There is a 95% probability that βⱼ lies in THIS specific interval." (βⱼ is a fixed
number, not random — it either is or is not in the interval.)
Practical use: If the 95% CI for β₁ is [0.05, 0.13], then we are confident the return to
education is between 5% and 13% — practically meaningful range.

5.3 Confidence Interval & Hypothesis Test Connection

There is a direct one-to-one relationship between confidence intervals and two-tailed hypothesis tests:
CI Approach t-Test Approach (Equivalent)
If H₀ value (e.g. 0) is OUTSIDE the 95% CI → |t| > t_{0.025} → Reject H₀ at 5% significance
Reject H₀
If H₀ value is INSIDE the 95% CI → Fail to reject |t| ≤ t_{0.025} → Fail to reject H₀
H₀

Worked Example: Confidence Interval

Model: log(wage) = β₀ + β₁educ + β₂exper + β₃tenure + u
β̂₁ = 0.092, se(β̂₁) = 0.007, df = 522, t_{0.025,522} ≈ 1.965
95% CI for β₁: 0.092 ± 1.965 × 0.007 = 0.092 ± 0.0138 =
[0.0782, 0.1058]
Interpretation: We are 95% confident the return to an extra year of education is between
7.82% and 10.58%, ceteris paribus.
Does this interval contain 0? No → Reject H₀: β₁ = 0 (confirms our t-test result)
6. THE F-TEST: TESTING MULTIPLE RESTRICTIONS

6.1 Why F-Tests?

The t-test tests one coefficient at a time. The F-test tests multiple hypotheses simultaneously. This
matters when:
• You want to test whether a group of variables is jointly significant
• Coefficients are correlated (multicollinearity makes individual t-tests misleading)
• You want to test whether the whole model explains anything at all

6.2 Setting Up the F-Test

The F-test compares two models:
Unrestricted The full model with all variables included. Has SSRᵤᵣ, Rᵤᵣ², df = n − k − 1
Restricted The model with the q restrictions imposed (variables dropped or coefficients set
to zero). Has SSRᵣ, Rᵣ², df = n − k − 1 + q

Key insight: Imposing restrictions always increases SSR (or decreases R²). The F-test asks: is the
increase in SSR too large to be due to chance?

F-Statistic F = [(SSRᵣ − SSRᵤᵣ)/q] / [SSRᵤᵣ/(n−k−1)]

q = number of restrictions tested; n−k−1 = df of unrestricted model. Under H₀, F ~ F(q, n−k−1).

R² Form of F-Statistic F = [(Rᵤᵣ² − Rᵣ²)/q] / [(1 − Rᵤᵣ²)/(n−k−1)]

Equivalent formula using R-squared. Convenient because R² is always reported. ONLY valid for
exclusion restrictions (same dependent variable in both models).

F-Distribution Properties
• Always ≥ 0 (can never be negative — it is a ratio of variances)
• Right-skewed
• Characterised by two parameters: numerator df = q, denominator df = n−k−1
• Reject H₀ only in the right tail (large F means restrictions are violated)

6.3 Types of F-Tests

Test 1: Joint Significance of a Subset of Variables
Null: β_{k-q+1} = β_{k-q+2} = ... = βₖ = 0 (q variables have no effect)
This is the most common F-test. Run the restricted model without those q variables, compute SSRᵣ,
then use the formula.
 Worked Example: Birth Weight Equation
bwght = β₀ + β₁cigs + β₂parity + β₃faminc + β₄motheduc + β₅fatheduc
+ u
H₀: β₄ = 0 AND β₅ = 0 (parents' education has no effect on birth weight)
q = 2 restrictions; n = 1191; k = 5; dfᵤᵣ = 1191 − 5 − 1 = 1185
Rᵤᵣ² = 0.0387 (full model) Rᵣ² = 0.0364 (restricted model, motheduc
& fatheduc excluded)
F = [(0.0387 − 0.0364)/2] / [(1 − 0.0387)/1185] = [0.00115] /
[0.000811] = 1.42
5% critical value: F(2, 1185) ≈ 3.0 → 1.42 < 3.0 → FAIL TO REJECT H₀
Conclusion: Parents' education is jointly insignificant in explaining birth weight, after
controlling for other factors. The p-value = 0.238.

Test 2: Overall Significance of the Regression

H₀: β₁ = β₂ = ... = βₖ = 0 (none of the explanatory variables explain y)
This tests the entire model. The restricted model is just ȳ (a constant).
F for Overall F = [R²/k] / [(1 − R²)/(n−k−1)]
Significance
R² = R-squared from the full model; k = number of regressors. Reported automatically in most
software.

6.4 Relationship Between t and F

t² = F for a Single Restriction
When testing a single coefficient (q = 1), the F-statistic equals the square of the t-statistic:
F(1, n−k−1) = [t(n−k−1)]²
They are equivalent for two-tailed tests. But for one-tailed tests, ONLY the t-test works (F
cannot detect direction). This is why we always use t-tests for single coefficients.

6.5 Important Warning: Multicollinearity and F-tests

Variables can individually have insignificant t-statistics but be jointly significant by F-test. This happens
due to multicollinearity:
• When two or more variables are highly correlated, individual t-tests have low power
• The F-test tests their combined explanatory power, bypassing the collinearity issue
• Example (Baseball salaries): HR per year and RBI per year were individually insignificant but
jointly significant at F = 9.55 >> 3.78 (1% critical value)
7. ERROR TYPES, SIGNIFICANCE LEVELS & POWER

7.1 Type I and Type II Errors

H₀ is TRUE (no real effect) H₀ is FALSE (real effect exists)

We REJECT H₀ TYPE I ERROR (False Positive) CORRECT DECISION (Power = 1

Probability = α (significance level) − β)
We FAIL TO REJECT CORRECT DECISION (Confidence TYPE II ERROR (False Negative)
H₀ = 1 − α) Probability = β

Trade-off: Lowering α (e.g. from 5% to 1%) reduces Type I errors but increases Type II errors (harder
to reject H₀). There is no free lunch.

7.2 Significance Levels

• α = 0.10 (10%): Lenient. Use when missing a real effect is costly (exploratory research).
• α = 0.05 (5%): Standard convention in economics and social sciences.
• α = 0.01 (1%): Strict. Use when falsely rejecting H₀ is very costly (e.g. medical studies).

Significance levels in published papers are usually denoted by stars:

* = significant at 10%, ** = significant at 5%, *** = significant at 1%

7.3 Power of a Test

Power = P(Reject H₀ | H₀ is false) = 1 − P(Type II Error)
A more powerful test is better. Power increases with:
• Larger sample size n → smaller se(β̂ⱼ) → easier to detect true effects
• Larger true effect size (bigger |βⱼ − 0|)
• Larger significance level α (at the cost of more Type I errors)
• Less multicollinearity (smaller Rⱼ²)
8. COMPREHENSIVE WORKED EXAMPLES

Example 1: Full t-Test Procedure (Two-Tailed)

Crime Deterrence: Effect of Conviction Probability on Arrests
Model: narr86 = β₀ + β₁pcnv + β₂ptime86 + β₃qemp86 + u (n =
2725)
narr86 = times arrested 1986; pcnv = proportion of prior arrests leading to conviction
Estimated model: narr86 = 0.712 − 0.150 pcnv − 0.034 ptime86 − 0.104
qemp86
se(β̂₁) = 0.041; df = 2725 − 3 − 1 = 2721
H₀: β₁ = 0 (conviction probability has no deterrence effect)
H₁: β₁ ≠ 0 (conviction probability affects arrest count)
t = −0.150 / 0.041 = −3.66
|−3.66| = 3.66 > 1.96 (5% critical value) → REJECT H₀
Also: 3.66 > 2.576 (1% critical value) → Significant at 1% level
Interpretation: A 0.50 increase in conviction probability reduces expected arrests by 0.075
(= 0.150 × 0.50). Higher conviction probability has a statistically significant deterrence
effect.

Example 2: Testing a Non-Zero Hypothesis

Testing Constant Returns to Scale in a Production Function
Model: log(output) = β₀ + β₁log(labour) + β₂log(capital) + u
Constant returns to scale requires β₁ + β₂ = 1 (doubling inputs doubles output)
H₀: β₁ + β₂ = 1 vs H₁: β₁ + β₂ ≠ 1
Suppose: β̂₁ = 0.63, β̂₂ = 0.30, se(β̂₁+β̂₂) = 0.09 (requires covariance
formula)
t = (0.63 + 0.30 − 1) / 0.09 = −0.07 / 0.09 = −0.78
|−0.78| < 1.96 → FAIL TO REJECT H₀. Constant returns to scale is supported by the
data.

Example 3: Full F-Test Procedure

Major League Baseball Salaries: Joint Test
Unrestricted: log(salary) = β₀ + β₁years + β₂gamesyr + β₃bavg +
β₄hrunsyr + β₅rbisyr + u
H₀: β₃ = β₄ = β₅ = 0 (batting avg, HR/year, RBI/year do NOT affect salary)
q = 3 restrictions; n = 353; k = 5; df = 353 − 5 − 1 = 347
SSRᵣ = 198.311 (years + gamesyr only)
SSRᵤᵣ = 183.186 (full model)
F = [(198.311 − 183.186)/3] / [183.186/347] = [5.042] / [0.528]
= 9.55
1% critical value: F(3, 347) ≈ 3.78 → 9.55 >> 3.78 → REJECT H₀ at 1%
Conclusion: Batting performance variables are jointly highly significant. Even though
individual t-statistics may be insignificant (due to multicollinearity between hrunsyr and
rbisyr), together they matter a great deal.
9. COMMON PITFALLS, EXAM TIPS & QUICK REFERENCE

9.1 Common Mistakes to Avoid

MISTAKE CORRECT APPROACH
"We accept H₀" when p-value is large "We fail to reject H₀" — absence of evidence is not
evidence of absence
Forgetting to use df = n−k−1 (not n−1) for t Count the intercept: k regressors + 1 intercept =
distribution k+1 parameters estimated. df = n − k − 1.
Using z critical values when sample is small (n Always use t distribution with correct df unless σ² is
< 60) known
Concluding economic significance from Always check the magnitude and sign of the
statistical significance alone coefficient alongside the p-value
Using R² form of F-test when the restricted Use SSR form of F-test when the dependent
model has a different dependent variable variable changes in the restricted model
Forgetting that F = t² only for a single restriction F(1, n−k−1) = [t(n−k−1)]². For multiple restrictions or
and two-tailed test one-tailed tests, use the appropriate statistic.

9.2 Quick Formula Reference Sheet

What You Need Formula
t-statistic (H₀: βⱼ=0) t = β̂ⱼ / se(β̂ⱼ) ~ t(n−k−1)
t-statistic (general) t = (β̂ⱼ − aⱼ) / se(β̂ⱼ)
Standard error se(β̂ⱼ) = √[σ̂² / (SSTⱼ(1−Rⱼ²))]
Estimated σ² σ̂² = SSR/(n−k−1) [degrees of freedom = n−k−1]
95% Confidence Interval β̂ⱼ ± t_{α/2, n−k−1} × se(β̂ⱼ)
F-statistic (SSR form) F = [(SSRᵣ−SSRᵤᵣ)/q] / [SSRᵤᵣ/(n−k−1)]
F-statistic (R² form) F = [(Rᵤᵣ²−Rᵣ²)/q] / [(1−Rᵤᵣ²)/(n−k−1)]
F for overall significance F = [R²/k] / [(1−R²)/(n−k−1)]
F = t² relationship F(1, n−k−1) = [t(n−k−1)]² [single restriction only]
p-value (two-tailed) p = 2 × P(T > |t|) where T ~ t(n−k−1)

9.3 Critical Values Quick Reference

Test / Distribution 10% 5% 1% Notes
z / N(0,1) — two-tailed 1.645 1.960 2.576 σ² known or n→∞
z / N(0,1) — one-tailed 1.282 1.645 2.326 Right tail
 t(30) — two-tailed 1.697 2.042 2.750 df = 30
t(60) — two-tailed 1.671 2.000 2.660 df = 60
F(1, 120) — right tail 2.75 3.92 6.85 q=1, df=120
F(2, ∞) — right tail 2.30 3.00 4.61 q=2, large n
F(3, ∞) — right tail 2.08 2.60 3.78 q=3, large n

9.4 Decision Rule Summary Diagram (in Words)

Two-Tailed Test at 5% Significance
Rejection Region | Acceptance Region |
Rejection Region
t < −1.96 | −1.96 ≤ t ≤ 1.96 |
t > 1.96
[Critical Region] | [Do Not Reject H₀] |
[Critical Region]
Left tail area = 2.5% Right tail area = 2.5%
Total rejection area = 5% = α

9.5 Your Lecture Topics — Mapping to This Guide

Topic from Your Notes Where to Find it in This Guide
Sampling Section 2: Sampling & Distributions
Testing for significance of a Sections 4.1–4.2: t-Test Procedure
coefficient
Approaches for denoting null Section 3.1: Types of Hypotheses
hypotheses
Types of hypotheses (H₀, H₁) Section 3.1–3.2: Hypothesis Framework
Test statistic (t-statistic, σ Sections 2.4, 4.1: t Distribution & t-test
unknown)
Critical regions vs Acceptance Section 3.3: Critical Regions & Rejection Rules
regions
Two-sided = two critical regions Sections 3.1, 3.3, 9.4: Two-Tailed Tests
z-test when σ is known Section 2.3: The Normal (z) Distribution
Confidence Interval approach Section 5: Confidence Intervals (entire section)
Significance level for 2-tailed α/2 Sections 5.1, 9.3: CI formula and critical values
Interpretation of Confidence Section 5.2: Interpretation
Interval
Decision: Reject / Fail to Reject Section 3.2: Step-by-Step Process + Section 9.1
10. MASTER CHECKLIST — CAN YOU DO ALL OF THIS?

Core Competencies for Mastery

DISTRIBUTIONS
7. Explain what the sampling distribution of β̂ⱼ is and why it is normally distributed
8. State when to use z vs t distributions and explain the difference
9. Interpret degrees of freedom and explain why small df gives fatter tails
t-TESTS
10. State H₀ and H₁ correctly for two-tailed and one-tailed tests
11. Compute the t-statistic for any null hypothesis H₀: βⱼ = aⱼ
12. Find the critical value from a t-table for given α and df
13. Apply the rejection rule and state the decision correctly (not 'accept H₀')
14. Interpret the result — both statistically and economically
CONFIDENCE INTERVALS
15. Construct a 95% (or any α%) confidence interval for βⱼ
16. Interpret a confidence interval correctly (frequentist interpretation)
17. Use the CI to draw the same conclusion as a two-tailed t-test
F-TESTS
18. Set up unrestricted and restricted models for any given set of exclusion restrictions
19. Compute the F-statistic using both the SSR and R² formulas
20. Know when to use the R² form vs the SSR form
21. Explain why jointly significant variables may have individually insignificant t-statistics
(multicollinearity)
p-VALUES & ERROR TYPES
22. Explain what a p-value means without saying 'probability H₀ is true'
23. Use p-values to draw conclusions at different significance levels
24. Explain Type I and Type II errors and the trade-off between them
25. Explain what statistical power means and what increases it

Based on: Wooldridge, J.M. (2012) Introductory Econometrics: A Modern Approach, 5th Edition, Chapter 4 — Multiple
Regression Analysis: Inference
Supplemented with standard econometrics references on hypothesis testing, distributions, and inference theory.