Related Experiment Videos
A simple and powerful test for autocorrelated errors in OLS intervention models.
1Department of Psychology, Western Michigan University, Kalamazoo 49008-5052, USA.
Psychological Reports
|October 12, 2000
Summary
A new statistical test for interrupted time-series regression models addresses weaknesses in existing methods. This approach offers accurate error independence evaluation without inconclusive results, improving time-series analysis.
Area of Science:
- Statistics
- Biostatistics
- Econometrics
Background:
- Interrupted time-series (ITS) regression models are crucial for evaluating interventions.
- A key assumption in ITS models is the independence of errors.
- Existing tests like Mood's runs test and Durbin-Watson (D-W) bounds test have limitations, including poor small sample performance and inconclusive results.
Purpose of the Study:
- To introduce a novel, simple-to-compute statistical test for evaluating the independence of errors in ITS models.
- To address the shortcomings of existing tests, specifically the lack of an inconclusive region and the provision of exact p-values.
Main Methods:
- A new statistical test for error independence in ITS models was developed.
- The test's properties were evaluated using Monte Carlo simulations.
- Comparisons were made against established tests, including Mood's runs test, D-W bounds test, and D-W beta test.
Main Results:
- The proposed test demonstrates good Type I error and power properties compared to existing methods.
- The test is simple to compute, requiring no specialized software.
- It provides an exact p-value and avoids the 'inconclusive' results often seen with the D-W bounds test.
Conclusions:
- The novel statistical test is a reliable and practical alternative for routine evaluation of error independence in ITS models.
- Its desirable properties, particularly the absence of an inconclusive region and exact p-value, enhance its utility in time-series analysis.