|
|
|
|
Benchmarking Bayesian
Hierarchical/Multilevel Models with Time Series Cross Sectional Data:
An Empirical Investigation |
|
| Abstract | Word Stems | Keywords | Association | Citation | Get this Document | Similar Titles |
|
|
Abstract:
|
Time series cross-sectional (TSCS) data has become
increasingly popular in political science research across the
subfields. It is well known that this data often exhibit serial
correlation, heteroscedasticity, and contemporaneous correlation
(Stimson 1985).
Earlier attempts at overcoming these problems involved running feasible
generalized least squares. Beck and Katz (1995) introduced a more
modern technique of combining OLS with panel-corrected standard errors
(PCSEs), after having corrected for serial correlation. In the past
several years, this estimator (justified by asymptotics in time), has
become very common in the literature.
More recently, Beck and Katz (2001) have considered “random
coefficients models” (RCMs) for TSCS data. They remarked on the
wholesale absence of RCMs for TSCS data in political science;
nevertheless, they speculated that fully Bayesian models may be a
superior methodology for this type of data.
One major exception to the absence of applications of RCMs to TSCS data
is the work of Western (1998), where he fits a Bayesian model to
economic growth in OECD countries. Western (1998) shows that his model
provides more accurate forecasts than other models, more accurate
estimates of time-series effects, and more realistic accounting of
uncertainty.
However, there are several important methodological oversights in his
analysis. First, he ignores the issue of serial correlation. Second, he
does not include an indicator for time. Failure to include a time
indicator leads to missing the common effects of particular years
across all cross-sectional units for that year. These “time shocks” are
a source of contemporaneous correlation. Third, he ignores the issue of
heteroscedasticity over time.
If one accounts for these issues, are his results generalizable? Recent
theoretical work by Western and Jackman (1994) and Steenbergen and
Jones (2002) suggests the answer could be yes. In particular, the
estimation of context or “group-level” effects—a particular strength of
Bayesian multilevel models—seems particularly well-suited to this data,
which often contain many time-invariant (“institutional”) predictors.
What remains to be done is an empirical investigation.
This paper will employ simulations in addition to multiple examples
from different substantive areas (state-level public opinion and
federal spending) to benchmark the performance of Bayesian
hierarchical/multilevel against OLS with PCSEs on this type of data. I
will analyze the extent to which multilevel models can do as well or
better than the older techniques through testing model fit and
comparing parameter and uncertainty estimates. Close attention will be
paid to the estimates of time-invariant predictors that represent
important contexts within which other factors play out. |
|
 | Convention | | Need a solution for abstract management? All Academic can help! Contact us today to find out how our system can help your annual meeting. |  | Submission - Custom fields, multiple submission types, tracks, audio visual, multiple upload formats, automatic conversion to pdf. |  | Review - Peer Review, Bulk reviewer assignment, bulk emails, ranking, z-score statistics, and multiple worksheets! |  | Reports - Many standard and custom reports generated while you wait. Print programs with participant indexes, event grids, and more! |  | Scheduling - Flexible and convenient grid scheduling within rooms and buildings. Conflict checking and advanced filtering. |  | Communication - Bulk email tools to help your administrators send reminders and responses. Use form letters, a message center, and much more! |  | Management - Search tools, duplicate people management, editing tools, submission transfers, many tools to manage a variety of conference management headaches! | | Click here for more information. |
|
|
Association:
Name: The Midwest Political Science Association
URL: http://www.indiana.edu/~mpsa/
|
Citation:
|
MLA Citation:
| Shor, Boris. "Benchmarking Bayesian
Hierarchical/Multilevel Models with Time Series Cross Sectional Data:
An Empirical Investigation" Paper presented at the annual meeting of the The Midwest Political Science Association, Palmer House Hilton, Chicago, Illinois, Apr 15, 2004 <Not Available>. 2008-10-10 <http://www.allacademic.com/meta/p83128_index.html> |
APA Citation:
| Shor, B. , 2004-04-15 "Benchmarking Bayesian
Hierarchical/Multilevel Models with Time Series Cross Sectional Data:
An Empirical Investigation" Paper presented at the annual meeting of the The Midwest Political Science Association, Palmer House Hilton, Chicago, Illinois <Not Available>. 2008-10-10 from http://www.allacademic.com/meta/p83128_index.html |
Publication Type: Conference Paper/Unpublished Manuscript
Review Method: Peer Reviewed
Abstract: Time series cross-sectional (TSCS) data has become
increasingly popular in political science research across the
subfields. It is well known that this data often exhibit serial
correlation, heteroscedasticity, and contemporaneous correlation
(Stimson 1985).
Earlier attempts at overcoming these problems involved running feasible
generalized least squares. Beck and Katz (1995) introduced a more
modern technique of combining OLS with panel-corrected standard errors
(PCSEs), after having corrected for serial correlation. In the past
several years, this estimator (justified by asymptotics in time), has
become very common in the literature.
More recently, Beck and Katz (2001) have considered “random
coefficients models” (RCMs) for TSCS data. They remarked on the
wholesale absence of RCMs for TSCS data in political science;
nevertheless, they speculated that fully Bayesian models may be a
superior methodology for this type of data.
One major exception to the absence of applications of RCMs to TSCS data
is the work of Western (1998), where he fits a Bayesian model to
economic growth in OECD countries. Western (1998) shows that his model
provides more accurate forecasts than other models, more accurate
estimates of time-series effects, and more realistic accounting of
uncertainty.
However, there are several important methodological oversights in his
analysis. First, he ignores the issue of serial correlation. Second, he
does not include an indicator for time. Failure to include a time
indicator leads to missing the common effects of particular years
across all cross-sectional units for that year. These “time shocks” are
a source of contemporaneous correlation. Third, he ignores the issue of
heteroscedasticity over time.
If one accounts for these issues, are his results generalizable? Recent
theoretical work by Western and Jackman (1994) and Steenbergen and
Jones (2002) suggests the answer could be yes. In particular, the
estimation of context or “group-level” effects—a particular strength of
Bayesian multilevel models—seems particularly well-suited to this data,
which often contain many time-invariant (“institutional”) predictors.
What remains to be done is an empirical investigation.
This paper will employ simulations in addition to multiple examples
from different substantive areas (state-level public opinion and
federal spending) to benchmark the performance of Bayesian
hierarchical/multilevel against OLS with PCSEs on this type of data. I
will analyze the extent to which multilevel models can do as well or
better than the older techniques through testing model fit and
comparing parameter and uncertainty estimates. Close attention will be
paid to the estimates of time-invariant predictors that represent
important contexts within which other factors play out. |
Get this Document:
Find this citation or document at one or all of these locations below. The links below may have the citation or the entire document for free or you may purchase access to the document. Clicking on these links will change the site you're on and empty your shopping cart.
Similar Titles:
What to do about Missing Data in Time Series Cross-Section Data
A Bayesian Multilevel Modeling Approach to Time Series Cross-Sectional Data
|
|
