Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 2 down vote favorite

I want to know if there is any literature about the following regression problem: $$ Y=X\beta +\epsilon$$ where $Y$ is unknown. But, i know $X$ and the OLS estimator of $\beta$ $$ \hat{\beta}=(X^\top X)^{-1}X^\top Y$$ because i know the value of $X^\top Y$ In fact i have an inverse problem of quadratic programming where i know the solution and i want the regression problem associated to this soluiton.

Thanks

share|improve this question
You just wrote down the estimator of $\beta$. What else do you want to know? – Aniko Feb 15 '12 at 20:11
i want to know if there any theories to find $Y$, to know the variance of $Y-X\hat{\beta}$ or the mean squared error associated. – david Feb 15 '12 at 20:20
Do you also know the sample covariance matrix of $\hat{\beta}$? – jbowman Feb 16 '12 at 14:44
how to know the covariance of the beta without $y$ ? – grant Feb 16 '12 at 18:41
Is Y a latent variable? If so, then something like structural equation modelling may be more appropriate. – Michelle Feb 16 '12 at 19:44
show 1 more comment

2 Answers

up vote 1 down vote

Perhaps 'calibration models' would give you some insights.

share|improve this answer

We're looking for long answers that provide some explanation and context. Don't just give a one-line answer: please explain why you're recommending it as a solution. Answers that don't explain anything will be deleted. See Good Subjective, Bad Subjective for more information.
up vote 1 down vote

There will be an infinite number of Y vectors that would lead to the same $\beta$ vector for the given $x$ matrix. One option for further studying the relationship would be to generate a random vector of "errors" from a normal distribution with mean exactly equal to 0 and orthogonal to $x\beta$ (you could do this by generating a random vector, regressing this onto $x\beta$ and taking the residuals) then adding those "errors" to $x\beta$. This will give you one of the $y$ vectors that would give the results you have. Now you can use this simulated $y$ with standard routines to find other values of interest. You can repeate this with multiple simulations, many of the standard results will be identical (provided the "errors" have the exact same variance).

share|improve this answer
this is very interesting. Do you know any paper about this ? – david Feb 16 '12 at 19:40
I don't know of a specific paper about this, but a decent book on linear models should contain the theory needed to justify it. – Greg Snow Feb 17 '12 at 18:25
thanks a lot for comments – david Feb 17 '12 at 23:24

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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