Testing a Nonlinear Model with a Time Series Data.

[Sungeun Chung]

Dear All,

I have a question about testing nonlinear model with a time series data.
I have a panel data: 95 cases, scores at 11 time points; there are  two
independent variables: Source Credibility (cred): high and low  credibility;
Message discrepancy (discrep): Three levels (7, 20, 40 years).  Position
change is the dependent variable - but 11 different time points.

My goal is to find a functional form for the data.  In particular, I am
testing the following nonlinear model:  y = a(1 - e^(-bt)), where y is
position change and t is time point (11 points).

I did an analysis using "nonlinear regression analysis" (see below).  For
this test, first I created 11 cases for each participant - Now I have 95 x
11 cases and one depedent variable.    Then, I created five dummy variables
for 6 different conditions by independent varaibles (2 x 3l d_dc1 thru
d_dc5) and did the following analysis.

NonLinear Regression.

MODEL PROGRAM b1=5 b2=1 b3 = 1 b4 = 2 b5 = 1 b6 = 6 b7=-.83 .

COMPUTE PRED_ = (b1 + b2 * d_dc1 + b3 * d_dc2+ b4 * d_dc3+ b5 * d_dc4+ b6 *
d_dc5) *(1 - EXP(b7 * time)).

CNLR pca

/PRED PRED_

/BOUNDS b1 >= 0; b7 <= 0

/SAVE PRED RESID

/CRITERIA STEPLIMIT 2 ISTEP 1E+20 .

It gave some results.  However, the data used are a panel.  There should be
some autocorrelations, which will give inaccurate standard errors.

There should be a way to test a nonlinear model with a panel data.  Can
anyone help me?

You help will be greatly appreciated.  If you need, I will send my data
(short version) and the syntax to you.

Thanks for your help in advance.

Sungeun
--
Sungeun Chung, Ph.D.

Assistant Professor
Department of Communication
Memorial Hall 303G
Western Illinois University
Macomb, IL 61455-1390
Tel. [phone]
E-mail: [hidden email]; [hidden email]