# Strange Chi2/ndf value

Dear All,
I have saved some data-sets in a tree. For each entry, there is an array of X and Y variables, which I save in a TGraph and then fit this graph by a linear function. Please note that all the numbers of the X and Y values are integer. For each fit, I am accumulating the fitted parameters in another TTree. At the end, I see the distribution of the chi2/ndf is peaked at 0.1 rather than 1. So the fits are understimated. So can any body tell me how to fix this type of problem? Thanks in advance.

if(gfit->Fit(“f1”,“NQO”)==0) {
chi2/ndf = f1->GetChisquare()/f1->GetNDF();
}

with regards,

Hi,

Which error are you using for each point ? If you are fitting TGraph you don;t have an error associated to the point, i.e. a default error of one is assumed, so looking at the value of Chi2/ndf does not make much sense, apart if the error in each point is really 1.

Best Regards

Lorenzo

Hello Lorenzo,
Thank you very much for your kind response. Yes, I have used TGraph to draw the data points. As I mesnitioned in my previous mail that all the events of the X and Y values are integers. So I have also tried to fit the graphs using the TGraphErrors class, where I have put Xerror = 1/\sqrt(12) and Yerror = 1/\sqrt(12) (i.e. standard error = 1/\sqrt(12)). But this method does not work for my case. Around 90% of the times, the linear function fails to fit the data.
However, the chi2/ndf distribution peaks at 1 position if I keep Yerror = 1/\sqrt(12) and Xerror = 0 for all the events in a TTree. So now the question arises, why the linear function and chi2/ndf are highly dependent upon the value of Xerror? Can I keep Xerror value = 0 in my code for all the events? Or I have keep its value something else. I am trying to analyze the data of a prototype detector, which consists of 12 layers of 1 \times 1 m^2 resistive plate chamber (RPC). Each RPC has 32 strips of X and Y planes.

with regards,