The centering/scaling means the xdata,ydata will be pre-normalized before any fitting is done. The doc is not specific about what the normalization is, but likely it is something like
xdata=(xdata-mean(xdata))/std(xdata)
ydata=(ydata-mean(ydata))/std(ydata)
As for which one you want, you want the one that gives the better fit. Look at the results and take the one that is better.
