Common mistake 1 with F-test, Wald & Chow

Testing a restriction you dreamt up after seeing the coefficients. That is data-mining and the p-value is meaningless.

Common mistake 2 with F-test, Wald & Chow

Using the Chow test with heteroskedastic errors - it assumes equal variances across the two regimes.

Common mistake 3 with F-test, Wald & Chow

Running the F-test after choosing the variables by looking at their t-statistics. The distribution it uses assumes the restriction was specified before seeing the data; testing what the data suggested inflates the statistic by an amount nobody can calculate.

Common mistake 4 with F-test, Wald & Chow

Concluding from a joint rejection that each variable matters individually. The F-test says the group carries information, which is compatible with any one of them being redundant given the others.

EconStatsCraft › F-test, Wald & Chow

F-test, Wald & Chow

Test restrictions on a fitted regression: joint significance of a group of variables, an arbitrary linear restriction, or a Chow test for a structural break.

What it needs

Dependent variable
All regressors in the model
Test these are jointly zero - Leave empty to test every slope at once.
Split the sample on

The intuition

Imposing a restriction can only make the fit worse. The F-statistic asks whether it got worse by more than you would expect from chance alone. Everything else - joint significance, Chow, testing a linear combination - is that one idea with a different restriction matrix R.

When to use it

After fitting a regression, when the question involves more than one coefficient - dropping a block of controls, testing constant returns to scale, or checking a structural break.

Where the estimator comes from

Write the restrictions as R\beta=q. Then F=\dfrac{(R\hat{\beta}-q)'\left[R(X'X)^{-1}R'\right]^{-1}(R\hat{\beta}-q)}{q\,\hat{\sigma}^2}, which is algebraically identical to comparing the restricted and unrestricted residual sums of squares, F=\dfrac{(\mathrm{SSR}_r-\mathrm{SSR}_u)/q}{\mathrm{SSR}_u/(n-k)}. The Chow test is exactly this with R imposing equality of every coefficient across two sub-samples.

The model

F=\frac{(\mathrm{SSR}_r-\mathrm{SSR}_u)/q}{\mathrm{SSR}_u/(n-k)}

What it assumes

Everything OLS assumes.
For an exact F-distribution, Normal errors - in large samples this stops mattering.
The Chow test assumes you know where the break is. If you do not, use Zivot–Andrews or CUSUM.
The restricted model is nested inside the unrestricted one - the same data, the same sample, one a special case of the other. Dropping a variable that has missing values changes the sample as well as the model, and the F-test then compares two things that are not nested at all.

Statistical properties

Exactly F-distributed under Normal errors; asymptotically χ²/q without them.
The Wald form uses only the unrestricted model; LR needs both; LM needs only the restricted one. All three are asymptotically equivalent, and in finite samples Wald ≥ LR ≥ LM.
The Wald statistic is NOT invariant to how you algebraically write a non-linear restriction - a known and awkward property.
Joint tests can reject when no individual t-test does. That is the classic multicollinearity signature, not a contradiction.

How to read the output

A large F means the restrictions make the fit much worse, so the data reject them.
Joint significance can reject even when no single coefficient is individually significant - that is the classic signature of multicollinearity.
Chow: rejecting means the two sub-samples have genuinely different relationships.
The F-statistic and its p-value answer one question: whether the restriction can be rejected. They say nothing about how large the effect is, and on a large sample a substantively trivial restriction will be rejected decisively.

A worked reading

F(2, 1195) = 36.53, p < 0.0001 for H₀: β_exper = β_expersq = 0. Experience matters jointly, and it matters through both terms at once: the linear term and the square are each estimated imprecisely because they move together, so reading either t-statistic alone understates the pair. The F-test asks the question the t-statistics cannot - whether experience belongs in the model at all - and answers it decisively.

Symptom → diagnosis → what to do

If you see this	It means	Do this
Joint F rejects, no single t does	Regressors are collinear	Keep them as a block; do not drop on individual t-values
Chow rejects	The relationship differs across regimes	Estimate separately, or add full interactions with a regime dummy
Chow with heteroskedasticity	The test assumes equal variances across regimes	Use a Wald test with robust errors instead
You chose the break date by looking at the data	The critical values are invalid	Use a sup-Wald / Quandt–Andrews test with the right distribution

Common mistakes

Testing a restriction you dreamt up after seeing the coefficients. That is data-mining and the p-value is meaningless.
Using the Chow test with heteroskedastic errors - it assumes equal variances across the two regimes.
Running the F-test after choosing the variables by looking at their t-statistics. The distribution it uses assumes the restriction was specified before seeing the data; testing what the data suggested inflates the statistic by an amount nobody can calculate.
Concluding from a joint rejection that each variable matters individually. The F-test says the group carries information, which is compatible with any one of them being redundant given the others.

Where to read this properly

Wooldridge (2019), Introductory Econometrics: A Modern Approach, 7th ed., Cengage, chapter 4
Chow (1960) - the structural-break test
Andrews (1993) - testing when the break date is unknown

Related Tests & breaks

Monte Carlo experiment · Structural breaks & stability

All analyses

Data & variables

Each example is simulated from a known model, so you can check whether the estimator recovers the truth - and nothing here is anyone's private data.

Filter variables

Drag a variable into a slot on the right - or just click it.

Choose an example dataset above to begin.

Choose a transformation - it opens in the data editor, with a live preview.

Apply to

Multiply by

Columns ctrl-click for several

Group by ctrl-click for several

Within groups

Statistic

Bands

How many

Cut points

Keep rows where

Value

and

Fill with

Value

Also mark which rows were filled

Rules

Base period = 100

Which part

Percent at each tail

Threshold

Survey weight optional

Direction

Names column

Values column

Largest first

Delete them, not just count them

Read this column as

Tidy the text

New name optional

EconStatsCraft rigorous, plain-spoken econometrics

Econometrics that explains itself.

Regression, panel data, time series and the whole battery of tests. Every result is followed by a verdict you can defend.

Load a dataset to see it here.

F-test, Wald & Chow

What it needs

The intuition

When to use it

Where the estimator comes from

The model

What it assumes

Statistical properties

How to read the output

A worked reading

Symptom → diagnosis → what to do

Common mistakes

Where to read this properly

See also

Related Tests & breaks

All analyses

Charts

Maps

Explore

Regression

Time series - one series

Time series - several series

Panel data

Multivariate statistics

Marketing research

Causal

Finance & risk

Actuarial

Tests & breaks

Machine learning

Econometrics that explains itself.