: On 18 Jun 2003 15:08:24 -0700, [EMAIL PROTECTED] (Dianne Worth)
: wrote:

:> My next question is this: 
:> Rather than multiple regression, I want to use structural equation 
:> modeling.  The sample is < 250, which I've been told is not large 
:> enough for LISREL.  The alternative software uses partial least 

The issue of sample size for SEM is rather confusing (to me, at least).  
The number 200 (or 250?) gets thrown around a lot, I think, because of an 
early paper by Boomsma which showed that there are convergence problems in 
a certain type of model when N < 200.  

There are many more things to consider, however, when determining the
appropriate sample size for SEM, not the least of which include the
distribution of the observed variables, the number of indicators per
latent variable, power considerations.  Below I've reproduced an excellent
brief summary of the literature that a PhD student in Norway posted to 
the SEM listserv some time back.  I'm sure you'll find more than this
if you search the SEMNET archives at 

http://bama.ua.edu/archives/semnet.html

Mike Babyak

>>> Joar Vittersx <[EMAIL PROTECTED]> 01/29/98 08:34am >>>
SEMNETTERS
 
Bentler & Chou (1987) suggested that a ratio as low as 5 subjects per
parameter would be sufficient if the data are normally  distributed:
"The ratio of sample size to the number of free parameters may be able
to go as low as 5:1 under normal and elliptical theory, especially when
there are many indicators of the latent variable and the associated
loadings are large." (p.173).  If the data are not normally distributed,
on the other hand, a sample size of 5000 subjects may sometimes be
needed (Hu & Bentler, 1995)   In general, Anderson and Gerbing argues
that a sample size  of 150 or more typically will be needed to obtain
parameter estimates that have standard errors small enough to be of
practical use (Anderson & Gerbing, 1988,  p. 415).  Boomsma (1982) found
that sample sizes of 100 to be accurate under ML estimation with normal
data. In a series of Monte Carlos studies with sample sizes between 150
and 1000, Finch, West & MacKinnon  (1997) found that estimates of model
parameters were generally unaffected by sample size. Barrett & Kline
(1981), using real data,  reported that a minimum of 50 subjects were
the minimum needed to reproduce the factor pattern of the Sixteen
Personality Factor Questionnaire. Finally, sample size as a function of
the number of variables was not found to be an important factor in
determining stability in structural equation modeling in a study by
Guadagnoli & Velicer (1988).  In this study, the sample sizes varied
from 50 to 1000 and the number of variables ranged from 36 to 144.
Component saturation and absolute sample size were found to be the most
important factors, and with high component saturation (i.e. factor
loadings of .80) solutions were stable across replicated samples
regardless of the number of indicators, even with as few as 50
participants.  With factor loadings of .60, a sample size of 150 should
be sufficient to obtain an accurate solution.
 
References
 Anderson, J. C., & Gerbing, D. W. (1988). Structural  Equation modeling
in practice : A review and recommended two-step approach. Psychological
Bulletin, 103(3), 411-423.
 
 Barrett, P. T., & Kline, P. (1981). The observation to variable ratio
in factor analysis. Personality Study and Group Behavior, 1, 23-33.
 
 Bentler, P. M., & Chou, C. (1987). Practical issues in structural
modelling. Socialogocal Methods and Research, 16, 78-117.
 
 Boomsma, A. (1982). The robustness of LISREL against smal sample size
in factor analysis models. In K. G. J�reskog & H. Wold (Eds.), Systems
under indirect observaion, Part 1 (pp. 149-173). Amsterdam:
North-Holland.
 
 Finch, J. F., West, S. G., & MacKinnon, D. P. (1997). Effects of sample
size and nonnormality on the estimation of mediated effects in latent
variable models. Structural Equation Modeling, 4(2), 87-107.
 
 Guadagnoli, E., & Velicer, W. F. (1988). Relation of sample size to the
stability of component patterns. Psychological Bulletin, 103(2),
265-275.
 
 Hu, L.-T., & Bentler, P. (1995). Evaluating model fit. In R. H. Hoyle
(Ed.), Structural Equation Modeling. Concepts, Issues, and Applications
. London: Sage.
 
 
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to