Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: Re: -mkspline- reference


From   "Lachenbruch, Peter" <Peter.Lachenbruch@oregonstate.edu>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Re: -mkspline- reference
Date   Wed, 6 May 2009 11:09:24 -0700

Because of the shape of this function, it is sometimes referred to as
"hockey stick" regression.  

Tony

Peter A. Lachenbruch
Department of Public Health
Oregon State University
Corvallis, OR 97330
Phone: 541-737-3832
FAX: 541-737-4001


-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Kit Baum
Sent: Wednesday, May 06, 2009 7:56 AM
To: statalist@hsphsun2.harvard.edu
Subject: st: Re: -mkspline- reference

<>
Section 7.4.1 of IMEUS (below) contains some of that material. It does  
not work from first principles, but the algebra involved with linear  
splines is very straightforward in terms of restricted least squares.  
For a linear spline, the function must equal the same value to the  
left and right of each knot point, as discussed in 7.4.1. Thus if we  
did not have -mkspline-, we could easily do what it is doing:

-----------------------------
webuse lutkepohl,clear
g t1 = 0
g t2 = 0
g a1 = tin(1960q1,1974q4)
g a2 = tin(1975q1,1982q4)
replace t1 = qtr if tin(1960q1,1974q4)
replace t2 = qtr if tin(1975q1,1982q4)
reg income a1 a2 t1 t2, nocons hascons
predict double ihat, xb
tsline ihat if tin(1960q1,1974q4) || tsline ihat if tin(1975q1,1982q4)
// 1982q4 is t = 60; apply continuity
constraint def 1 a1 + t1*60 = a2 + t2*60
cnsreg income a1 a2 t1 t2, nocons c(1)
predict double ihat2, xb
tsline ihat2 if tin(1960q1,1974q4) || tsline ihat2 if tin(1975q1,1982q4)
--------------------------------

Compare the two graphs. Notice that the constrained function is  
continuous (but not differenttiable) at the single knot point.  (We  
could easily add additional knot points to this function).  If we  
imposed that the derivative of the function was equal to the left and  
right, we'd have a quadratic spline. If we imposed that both first and  
second derivatives were equal to the left and right, we'd have a cubic  
spline. But linear splines are just predicted values that obey  
continuity over the variable at each knot point.

Kit Baum   |   Boston College Economics & DIW Berlin   |
http://ideas.repec.org/e/pba1.html
                               An Introduction to Stata Programming   
|   http://www.stata-press.com/books/isp.html
    An Introduction to Modern Econometrics Using Stata  |
http://www.stata-press.com/books/imeus.html



On May 6, 2009, at 02:33 , statalist-digest wrote:

> Thanks.  Actually, -mkspline- computes *linear* splines. For instance,
> the following code:
>
> . mkspline  Age1  44  Age2 = Age
>
> . regress Y Age1 Age2
>
> fits a continuous, piecewise linear model of E(Y|Age) for which the
> slope is permitted to change abruptly at Age=44.
>
> Thus I am looking for a reference that specifically explains what
> algorithm and formulas are implemented in -mkspline-, and perhaps also
> discusses uses of linear splines and gives examples where they are
> used to good effect.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index