Statistical Stew
Thread Score:
Page 1 of 1
Thread Actions

9/22/2017 at 10:49:31 PM GMT
Posts: 12
Statistical Stew
I'm reviewing a TRB paper. It's a zero-inflated negative binomial model of transit trips within block groups nested within seven cities, with some multi-colinearity issues.
 
I'm not sure what to suggest they fix, it what is most important to fix. Either SEM would fix the multi-colinearity issue, with PCI being simpler to implement. 
 
But looking at the coefficients, which are all over the map, some sort of multilevel model seems necessary. But I'm not sure how that interacts (statistically speaking) with ZINB model. 
 
Could meta-regression techniques be used to combine coefficients? For generalizing their results, MLM seems to be the most important thing. I'm not sure what to suggest to the authors. 
 


9/25/2017 at 6:53:36 PM GMT
Posts: 1
Kind of hard to respond without seeing it :-)

But seems like you have answered your question already in some ways. If the model specification does not match the underlying causal logic, then they need to return to conceptual model and use either SEM or MLM based on theory and causal process suggested...unless it is an early exploratory study without a well defined theoretical framework? Or alternatively consider a sequential approach to modeling that better reflects order -
hierarchy of causality....

If it appears to be a kitchen sink model (throw everything in and see what pops up significant) that can only be justified in the early stages of new phenomena - exploratory research (at least thats my two cents) Cheers!


9/25/2017 at 7:46:51 PM GMT
Posts: 1
Since they're using a small geography--block groups--spatial autocorrelation may also be an issue. Local Indicators of Spatial Autocorrelation (LISA) at the level of each city, then all together could be helpful. Also, since block groups are often separated by street centerlines, transit trip characteristics may get inappropriately split between block groups, muddling effects. Geographically weighted regression may help address autocorrelation, but has several requirements.
Anselin, L. (1995). Local Indicators of Spatial Association-LISA. Geographical Analysis, 27(2).
Anselin, L., Syabri, I., & Kho, Y. (2006). GeoDa: An Introduction to Spatial Data Analysis. Geographical Analysis, 38(1), 5–22. https://doi.org/10.1111/j.0016-7363.2005.00671.x
Leung, Y., Mei, C. L., & Zhang, W. X. (2000). Statistical tests for spatial nonstationarity based on the geographically weighted regression model. Environment and Planning A, 32(1), 9-32.


Mission

The Association of Collegiate Schools of Planning promotes education, research, service and outreach in the United States and throughout the world by seeking to:

  • recognize diverse needs and interests in planning;
  • improve and enhance the accreditation process, and;
  • strengthen the role of planning education in colleges and universities through publications, conferences, and community engagement;
  • extend planning beyond the classroom into the world of practice.

Connect

2927 Kerry Forest Parkway • Tallahassee FL 32309