Detecting fake results


New Member
Hello everyone,

I am currently working together with a fellow student on analyzing what factors affect profitability in banking, by using a multivariate regression and OLS.

I am however suspicious that he is faking the results he is getting, as they seem to good to be true. He always have excuses for not giving me the data is he using etc. So I am wondering if there is any way find out if he is faking his results, by just analyzing the regression output from eviews that he sent me? My suspicion is that he is just replacing the coefficients and p values with what he wants it to be...

This may seem like a wierd question, but thanks in advance.



Omega Contributor
Well if you can get the normal outputted table from say the OLS, you can probably check to see if all of the numbers add up. I would imagine it would take quite a bit of work to substitute the right coefficients, SE, and test-statistics for the respective df, to get everything to work out. Feel free to post an example of the output.


Smelly poop man with doo doo pants.
you could also just ask him for the covariance matrix of the data?

not many people know you can obtain the regression coefficients from the covariance matrix and i think it would take your colleague quite a bit of time to modify every single one of the elements of said matrix to make sure they match whatever coefficients he's given you.


New Member
Thanks for the quick answers!
Here are the OLS output and correlation matrix.

The variable I am concerned about is the "fincrisis"


Smelly poop man with doo doo pants.
just for kicks and giggles i tried to obtain the test statistic and p-value of 3 regression coefficients of the variables you posted from the outputs (including fincrisis) and i was able to reproduce the results from your table. so, like hlsmith mentioned, if he really is faking it he's also taking the time to not only alter the regression coefficients but he's also changing the standard error and re-creating the test statistic and the p-value to make sure everything agrees.

with the info you provided you still cannot obtain the regression coefficients from the covariance matrix because you didn't provide a covariance matrix. you'd need the standard deviations of the the variables to transform the correlation into a covariance matrix and then see if what you get from there matches the unstandardized regression coefficients.

what you could do, however, is use the formula for the multiple correlation coefficient that relies only on the correlation matrix and see if it matches the multiple correlation coefficient/R-squared value of 0.4672 reported.

but, at this point, i think we have one of 3 scenarios:

(1) you have an EXTREMELY careful cheater who's taking the time to fake even the standard errors of its analyses and make sure everything in the results table agrees.

(2) your colleague is tampering with the data from the beginning (in which case you'll never be able to figure it out with the info you have now).

OR... the unfathomable...

(3) he's not cheating.

i now feel compelled to summon one of my most powerful didactic tools!



Omega Contributor
Yes, I was going to add that the falsification could be happening at the data level, but he would have to make sure he was systematically changing fincrisis given the values of Y. Which wouldn't be too hard to do. You could ask for the descriptive statistics for the variables and see if they make since.

It also seems like they may be controlling by clusters, so is this a multilevel regression model?