How to correctly perform a sideband fit

Dear all,

I am having a very hard time to find the correct way of fitting the sideband of a distribution with a straight line.

I have found countless bugs, the last one if the bug concerning RooProdPdf therefore I am wondering if you have a good solution for my problem.

After many attempts I am now running this code which I though it was working but I see now that it print different result for the fit if someone if excluding or not the data in RooDataSet in the central window. This of course should not affect the fit which is performed only in the sideband. Have you got any idea or where I am hitting a bug? Thanks

{

  double minL = 4829.24;
  double MAXL = 4887.06;
  double minR = 5060.54;
  double MAXR = 5118.36;

  RooRealVar *mHNL = new RooRealVar("Lambda0_M","m_{HNL}",minL,MAXR,"MeV/c^{2}");
  mHNL->setRange("R",minL,MAXR) ;
  mHNL->setRange("R1",minL,MAXL) ;
  mHNL->setRange("R2",minR,MAXR) ;

  RooRealVar leftEdge  ("lE","lE",minL);
  RooRealVar rightEdge ("rE","rE",MAXL);
  RooRealVar leftEdge1 ("lE1","lE1",minR);
  RooRealVar rightEdge1("rE1","rE1",MAXR);

  RooGenericPdf stepFuncPDF("stepFuncPDF", "stepFuncPDF",
      "((@0 >= @1) && (@0 < @2)) || ((@0 >= @3) && (@0 < @4))",
      RooArgList(*mHNL,leftEdge,rightEdge,leftEdge1,rightEdge1));

  RooRealVar *a0 = new RooRealVar("a0", "coefficient of x^1 term",-1.,1.);
  RooChebychev *p0 = new RooChebychev("p0","p0",*mHNL,RooArgSet(*a0));

  RooProdPdf  *pp = new RooProdPdf("pp","pp",RooArgSet(stepFuncPDF,*p0));

  RooDataSet* mHNL_dataset = pp->generate(*mHNL,100) ;

  // --- Fit
  RooFitResult* fitep0 = pp->fitTo(*mHNL_dataset,RooFit::Optimize(1),RooFit::Range("R1,R2"),RooFit::Save());
   cout<<"  +++  Fit results  +++  "<<endl;
  fitep0->Print();
  cout<<"Fit status= "<<fitep0->status()<<"; covQual= "<<fitep0->covQual()<<endl;

  // --- Plot
  RooPlot* frame = mHNL->frame() ;
  mHNL_dataset->plotOn(frame, RooFit::DrawOption("Zp")) ;
  pp->plotOn(frame);
  frame->Draw();

}

HI

Are you sure you see a real effect and not a statistical one, sure to the fact that the events generated might be different ?
If you are sure this is not the case, can you please explains exactly in the macro how you obtain the different results

Best Regards

Lorenzo