Date:         Mon, 23 Oct 2006 20:24:58 -0300
Reply-To:     rdporto1 <rdporto1@TERRA.COM.BR>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         rdporto1 <rdporto1@TERRA.COM.BR>
Subject:      Re: LOESS and LOWESS
Comments: To: "ted.barker" <ted.barker@GMAIL.COM>
Content-Type: text/plain; charset=iso-8859-1

Ted,

I don't think PROC LOESS generates diagnostic plots. Anyway,
for detecting non linearities, you can try other approachs like
splines or wavelets. Maybe one of them can handle binary outcomes.

If you have some idea of the non linearities, you can also
try some generalized linear (or additive) model and use
their diagnostic plots. However, this is not an easy task.

HTH,

Rogerio Porto.

---------- Cabeçalho original -----------

De: "SAS(r) Discussion" SAS-L@LISTSERV.UGA.EDU
Para: SAS-L@LISTSERV.UGA.EDU
Cópia:
Data: Mon, 23 Oct 2006 12:52:20 +0100
Assunto: LOESS and LOWESS

> Hello:
>
> I am trying to examine measurement invariance of a factor at several
> adjacent time points. I believe there may be non-linear relationships. The
> indicators are binary.
>
> I have been reading Bauer, J. (2005).The role of nonlinear
> factor-to-indicator relationships in tests of measurement equivalance,
> Psychological Methods, 10, 305-316. Bauer suggested using PROC LOESS in a
> two-step strategy to diagnose non-linear effects in a factor model (via
> diganostic plots).
>
> I am aware via an earlier post in SAS-L that LOESS does not accomodate
> binary outcomes.
>
> Question: can the LOWESS macro be used to generate diagnostic plots for
> examining non-linear effects in factor models based on binary indicators?
> That is, can binary DVs be used in LOWESS?
>
> TIA & Best regards
>