[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
Jeffrey Simons <jsimons@usd.edu> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
st: Re: interactions, slopes, and standard errors |

Date |
Sat, 27 Mar 2004 08:31:53 -0600 |

Has anyone written a general program to calculate and display the estimated slope and associated error of Y on x1 at various values of x2, where Y = b0 + b1x1 + b2x2 + b3x1x2 Last fall Scott Merryman posted a nice program (which I copy in, below) which takes care of this at the mean of x2. Thanks, Scott!, We could also be interested in doing it for other values. Chip Hunter I have written a program that does this and makes graphs or the simple slopes at various multiples of the SD of x2. It will do it for several forms of interactions, e.g., 2-way, 3-way, quadratic effects etc. The help file is below. I'd be happy to make it available to anyone interested but thought it could still use some more refinement. Jeffrey Simons {smcl} {.-} help for {cmd:sslope} {.-} {title:Simple Slope Calculation for Linear Regression} {p 8 8} {cmd:sslope} {it:varlist} [{cmd:if} {it:exp}] [{cmd:in} {it:range}] {cmd:,} [{cmd:i}{cmd:(}{it:x z xz}{cmd:)}] [{cmd:i}{cmd:(}{it:x z w xz xw zw xzw}{cmd:)}] [{cmd:i}{cmd:(}{it:x x-squared}{cmd:)}] [{cmd:i}{cmd:(}{it:x z x-squared xz}{cmd:)}] [{cmd:i}{cmd:(}{it:z x x-squared xz}{cmd:)}] [{cmd:sd}{cmd:(}{it:#}{cmd:)}] [{cmd:Graph}] or [{cmd:go(}{cmd:}graph options{cmd:)}] [{cmd:Fits}] {p 8 27} {it:varlist} = regression equation with {ul:centered variables} including all interactions {p 8 8} option {cmd:i} must include either {cmd:(x z xz)} (2-way interaction), gives slope of criterion variable on x at levels of z (x z xz are variable names) {p 8 8} OR {p 8 8} {cmd:(x z w xz xw zw xzw)} (3-way interaction), gives slope of criterion variable on x at levels of z and w {p 8 8} OR {p 8 8} {cmd:(x x-squared)} (quadratic), gives slope of criterion variable on x at levels of x. Also calculates the minimum or maximum of the curve. {p 8 8} OR {p 8 8} {cmd:(x z x-squared xz)} (quadratic plus linear interaction), gives slope of criterion variable on x at levels of x and z. Also calculates the minumum or maximum of the curves. {p 8 8} OR {p 8 8} {cmd:(z x x-squared xz)} (quadratic plus linear interaction), gives slope of criterion variable on z at levels of x. {title:Description} {p} This calculates simple slopes for interactions between continuous variables in linear regression analysis. For example, the slope of the criterion on x conditional upon z. All continuous variables in the regression must be centered at the mean. The simple slope of interest must be specified in {cmd:i}. This must be of the form x(slope of interest) z(moderating variable) xz(the interaction between the two). Hence, {cmd:i}(x z xz) provides a different result than {cmd:i}(z x xz). Both are correct, one is the slope of y on x conditional upon z and the other, the slope of y on z conditional upon x. However, {cmd:i}(x xz z) will produce incorrect results. Similarly, for a three-way interaction, {cmd:i} must be in the form of (x z w xz xw zw xzw) for the slope of x conditional upon z and w. {ul:Important}, the order i(x z w xz {it:wz} xw xzw) will not produce the correct results. The order is x (slope of interest) z w (first order terms of the moderators) x^2 (x quadratic term) xz xw (2-way interactions with x) zw (the 2-way interaction of the moderators without x) xzw (the three way interaction). The order of variables in varlist does not matter but calculations of the simple slopes depend upon the specified order in {cmd:i}. Simple slopes associated with both an interaction and a quadratic term (i.e., y on x when there are both xz and x-squared terms) are not supported at this time. The calculations are based upon the formulas provided in Aiken & West (1991). {title:Options} {p} {cmd:sd}{cmd:(#)} specifies the conditional levels of interest of z and/or w, in sd units, default = 1.0. {p} {cmd:Graph} requests a graph of the conditional slopes. Continuous variables not included in the interaction are held constant at mean=0, if there are dichotomous variables in the model, this option creates graphs for a category=0, if there is not a category=0, the intercepts will be invalid. {p} {cmd:go} requests a graph as above, but accepts all standard graph options for {cmd:scatter}. Choose either {cmd:Graph} for the default graph OR {cmd:go} and define graph options. {p} {cmd:Fits} requests the conditional fitted values be saved. This is included to facilitate making additional graphs. {p} {title:Examples} sslope y x z xz, i(x z xz) sslope y x z xz, i(x z xz) sd(2) sslope y x z w xz xw zw xzw, i(x z w xz xw zw xzw) sslope y q x z w xz, i(x z xz) sslope y x x-squared, i(x x-squared) graph sslope y x z x-squared, i(x x-squared) graph fits sslope y x q z xz x-squared, i(x z x-squared xz) go(sort c(. l l l) m(o i i i) mlabel(id)) fits {bf} {dialog sslope:sslope dialog box} {bf}Author {sf}Jeffrey S. Simons - jsimons@usd.edu {bf}Also see {help mcenter} * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

**Follow-Ups**:**st: Finding log10(hessian condition number of a regression matrix)***From:*"Clive Nicholas" <Clive.Nicholas@newcastle.ac.uk>

- Prev by Date:
**st: RE: marginal effects** - Next by Date:
**st: RE: merging variable labels** - Previous by thread:
**Re: st: Emacs Speakes Statistics** - Next by thread:
**st: Finding log10(hessian condition number of a regression matrix)** - Index(es):

© Copyright 1996–2016 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |