In going over some example problems from class notes, I came across the following problem:

"Suppose we have a regression model $y=\beta_0+\beta_1x_1+\beta_2x_2+\beta_3x_4+\beta_4x_4+\epsilon$. We wish to test the null hypothesis that $H_0: \beta_1=\beta_2$. Express the hypothesis in terms of $A\beta=c$, provide $A$ and $c$ which yield the correct formulation of $H_0$. What is the rank of $A$?"

So, unless I am misunderstanding the problem, I would go about answering it thusly:

1) Restate the null hypothesis as $H_0: \beta_1-\beta_2=0$.

2) Set $c=0$.

3) Then, for $A\beta=c$, we have $A=[0,1,-1,0,0]$ and (no idea how to typeset a column vector) $\beta'=[\beta_0,\beta_1,\beta_2,\beta_3,\beta_4]$. Do the vector multiplication out, and you end up with $\beta_1-\beta_2=0$, which is our null hypothesis.

But, then, how do I answer the last part of the question? The rank of $A$? I've never even heard anyone talk about rank in the context of a vector. Did I do something wrong here and $A$ is supposed to be some $m$ x $n$ matrix for which I could calculate a rank? Or was this just a bad question (the professor in question is notorious for asking questions that don't make any sense)?

  • $\begingroup$ It looks to me like $A$ is a $1\times 5$ matrix. $\endgroup$
    – whuber
    Commented Aug 6, 2014 at 14:08
  • $\begingroup$ This may help you resolve one part of your confusion - en.wikipedia.org/wiki/… $\endgroup$
    – Glen_b
    Commented Aug 6, 2014 at 14:11

1 Answer 1


Your procedure is correct.

The rank can be very well interpreted for a single vector as well: the (row) rank will be the number of linearly independent rows. Given that a row vector has a single row it translates to whether the row will be linearly independent itself. Considering now the the definition of linear dependence (''has a linear combination with not every coefficient being zero that is the zero vector'') applied to a single vector (''multiplied with a non-zero scalar it is the zero vector'') it is immediately obvious that a non-zero vector is linearly independent (there is no scalar with which we can multiply a non-zero vector and the result will be the zero vector) and a zero vector is linearly dependent (the multiplier can be any scalar).

To sum up: the rank is 0 if the vector is the zero vector, 1 otherwise. In your case, the rank is thus 1.

In that specific context, the question makes some sense (at least generally speaking...) as it is customary to check prior to such parameter-restricion testings whether matrix $A$ is of full row rank so that the equations are not redundant. (See Hayashi Econometrics, p. 39.)

  • $\begingroup$ Thanks Tamas! That is more or less the answer I would have guessed, I just had no logical explanation for why I would choose it. $\endgroup$ Commented Aug 6, 2014 at 15:32
  • $\begingroup$ @RyanSimmons: You're welcome! $\endgroup$ Commented Aug 6, 2014 at 15:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.