|
Activity Number:
|
514
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 10, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Business and Economics Statistics Section
|
| Abstract - #307433 |
|
Title:
|
How Far to Asymptopia? Errors and Regressors with Realistically Asymmetric Distributions Generate Unreliable t-Statistics in Large Finite Samples
|
|
Author(s):
|
Robert McClelland*+ and Elliot Williams
|
|
Companies:
|
Bureau of Labor Statistics and Bureau of Labor Statistics
|
|
Address:
|
Suite 3105, Washington, DC, 20212,
|
|
Keywords:
|
central limit theorem ; Berry Esseen
|
|
Abstract:
|
Various central limit theorems guarantee that the distribution of t-statistics for significance of regression coefficients is asymptotically Normal under regularity conditions. Without violating the assumptions of the CLTs, however, characteristics of the data can dramatically slow the convergence of the finite-sample distributions to their Normal limiting distribution. Here we show that asymmetrically distributed regressors and errors can significantly slow convergence of the tail quantiles of the finite sample distribution of the t-statistic to their Normal limiting values. Particularly, we find that the asymptotic Normality assumption is misleading for inference at the 5% level with sample sizes from 10,000 to 100,000 when applied to data with skewnesses typical of economic wage and income data.
|
