The residual distribution should be normal with mean zero if possible. However, I have a situation where after nonlinear least squares, I get a residual distribution which is not normal. Kind of uniform distribution or highly skewed or else. Now, I have 4 parameters and 8 data points. Is it happening because the number of data points is small and there are simple not enough residuals to form a normal distribution. Or is it because of a fundamental incorrectness of the model. If it is the earlier then is there a minimum number of data points necessary for robust analysis(depending on the number of parameters).
8 datapoints for the estimation of 4 parameters is too little. In order tot estimate 4 parameters i would suggest you use at least 20 datapoints! That's a bit a rule of thumb for a minimum number. I've seen others propose something between 5-15 as minimum number of datapoints per variable
Last edited by Eyecatcher; 10-30-2012 at 04:24 AM.