R help - May 2008 - interpreting significance of path coefficients from sem() output

Hi there,

Quick question about the output from the sem() function in the library 
of the same name.

If I am getting probabilities >0.05 for some of my estimates of path 
coefficients, I'm assuming the interpretation here is that the 
coefficient is not significantly different from zero, correct? In that 
case, might it make sense that I should disregard path coefficients 
between variables where the probability is greater than 0.05? In which 
case, would it make further sense to remove those particular links from 
the specify.model() command and re-run the analysis, excluding those 
rows which lacked significance in the previous attempt?

Given that in just about every other example I've been able to dig up 
where this method is employed (including other datasets I am working 
with), the probabilities of the path coefficients are almost always well 
below 0.05, it makes me suspect that what I am observing may simply 
result from the fact that I'm trying to fit a path analysis among 5 
variables (4 predictors, 1 criterion) based on only 18 observations, 
admittedly a small sample size and perhaps an overly ambitious approach 
to analyzing so few data.

Last, I'm convinced that I'm using the code correctly as I was able to 
successfully reproduce an example in Quinn and Keough (2002) before I 
turned the code onto my own data.

Looking forward to any thoughts or comments.

Cheers,

Mike

-- 
Michael D. Rennie
Ph.D. Candidate
University of Toronto at Mississauga
3359 Missisagua Rd. N. 
Mississauga, ON L5L 1C6
Ph: 905-828-5452 Fax: 905-828-3792
www.utm.utoronto.ca/~w3rennie

Michael,

As a general rule of thumb (I believe this is in Jim Grace's book, if not
others) one should use 10-20 observations per variable.  If you have 5
variables, and 18 observations, you should probably be a bit suspect of your
results.  That said, if some of your paths are indeed non-significant, well,
they might be!  Have you tried an alternate model with those paths set to 0? 
You can them compare the two models in a variety of ways (LR tests, compare
BIC values, etc).

-Jarrett


Michael Rennie-2 wrote:
> 
> 
> Hi there,
> 
> Quick question about the output from the sem() function in the library 
> of the same name.
> 
> If I am getting probabilities >0.05 for some of my estimates of path 
> coefficients, I'm assuming the interpretation here is that the 
> coefficient is not significantly different from zero, correct? In that 
> case, might it make sense that I should disregard path coefficients 
> between variables where the probability is greater than 0.05? In which 
> case, would it make further sense to remove those particular links from 
> the specify.model() command and re-run the analysis, excluding those 
> rows which lacked significance in the previous attempt?
> 
> Given that in just about every other example I've been able to dig up 
> where this method is employed (including other datasets I am working 
> with), the probabilities of the path coefficients are almost always well 
> below 0.05, it makes me suspect that what I am observing may simply 
> result from the fact that I'm trying to fit a path analysis among 5 
> variables (4 predictors, 1 criterion) based on only 18 observations, 
> admittedly a small sample size and perhaps an overly ambitious approach 
> to analyzing so few data.
> 
> Last, I'm convinced that I'm using the code correctly as I was able
to
> successfully reproduce an example in Quinn and Keough (2002) before I 
> turned the code onto my own data.
> 
> 
-- 
View this message in context:
http://www.nabble.com/interpreting-significance-of-path-coefficients-from-sem%28%29-output-tp17096769p17096869.html
Sent from the R help mailing list archive at Nabble.com.