Estimating a breakpoint statistically for a piecewise regression

[Nick Cain]

Hello Stata List,

I'm interested in estimating a breakpoint for a time series data set. I know that the Chow Test can be used to assess whether a known breakpoint is statistically significant and I found a short UCLA stats article (http://www.ats.ucla.edu/stat/stata/f...imal_knots.htm) on splitting a piecewise regression, but I'd like to find a routine that can determine the proper breakpoint in an unsupervised fashion. Is this possible?

I've seen a few posts about the nlhockey routine, but the online help info I've been able to find is rather sparse. Does any know of a more detailed demonstration of this routine or a different approach I can try?

Thanks in advance!

-nick

[Clyde Schechter]

I'm not aware of any demonstration materials for -nlhockey-, but it's very simple to use. If you download it and read the help file, you should have no problem with it.

[Michael N Mitchell]

My book Interpreting and Visualizing Regression Models using Stata includes substantial coverage of piecewise models, including fitting models with an unknown knot (see Chapter 4). It also covers the interaction of a categorical variable by a continuous variable modeled in a piecewise manner (see Chapter 12). You can see the table of contents at http://www.stata-press.com/books/int...ession-models/ .

I hope this is useful,

Michael N Mitchell