Non normal residuals

Hi everyone. Im rookie in this forum and also in regression analysis so i hope there I will be able find the help :)
I created OLS multivariable regression model and almost all looks pretty good. All variables are significant, R-squared - also good, near to 0.8. It was a little problem with heteroskedaticity, but with logarithms i fixed it. But another problem still exists. Residuals of my model are still non normal and I dont know what i need to do. I checked correlation of residuals and other variables and there are no correlation.
Is it a very big problem or I can ignore it? Maybe somebody of you could help me to fixed it.
Thanks for advice and help!

Right here is the graph of my residuals


Less is more. Stay pure. Stay poor.
These residuals do not look too off considering how skewed they can be at times. I would also generate a Q-Q plot and if the residuals following the 45 degree line without too much deviation than you should be fine.

I am guessing you have a large sample, which tests for normality can tend to fail in some of those circumstances, but data displays can be close enough for the normality assumption.


No cake for spunky
Real world data is rarely perfectly normal and that looks pretty close. As hlsmith said look at a qq plot and if it is close than you should be ok. Normality only effects the test of significance not the parameter estimates. There are more powerful transformations than logging if you really want to do them. Look at the tukey ladder of power for that.

You might also request a skew and kurtosis test which all commercial statistical software produces. There are more advanced methods to look for outliers and high leverage points - but if you are new to regression you might not want to do that :p Outliers can be detected by looking at studentized values beyond 2 (some say 3) while leverage can be detected by some variation of Cooks's D (there are many variations of this). Other tests generate how much deleting one point will move the regression line (DFBETA).

But again that is going well beyond what most would do to start.