Adaptive Design Optimization for Model Discrimination under Model Misspecification
Main Author: | |
---|---|
Language: | English |
Published: |
The Ohio State University / OhioLINK
2012
|
Subjects: | |
Online Access: | http://rave.ohiolink.edu/etdc/view?acc_num=osu1347034580 |
id |
ndltd-OhioLink-oai-etd.ohiolink.edu-osu1347034580 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-OhioLink-oai-etd.ohiolink.edu-osu13470345802021-08-03T06:06:31Z Adaptive Design Optimization for Model Discrimination under Model Misspecification Sun, Yinghao Psychology <p>An experiment in cognitive science is often conducted with the goal of discriminating multiple competing models of a cognitive process. Adaptive design optimization (ADO) is a statistical methodology for selecting the values of the critical design variables (e.g., presentation schedule, stimulus structure) to present on each experimental trial based on responses from the preceding trials such that the chosen values are most informative in differentiating between models under consideration. Prior to applying ADO, a set of candidate models to be discriminated must be specified. Implicit in ADO is the essential assumption that one of the candidate models generates the data, the assumption that may be violated in practice due to model misspecification in which by definition data are generated by none of the candidate models. Even though under model misspecification, identifying the exact form of the data-generating model may not be possible, in present thesis we asked whether ADO model discrimination would nevertheless choose the candidate model that was most similar to the data-generating model in predictions.</p><p>Two candidate models of retention memory, power (POW) and exponential (EXP), were compared in two simulations in which the data were generated by a third, different model, that is, when model misspecification happened. In the first, the candidate model predictions were distinct and thus the model discriminability was high. The data were generated from either of two hyperbolic (HYP) models instead of from any of the candidate models. We found that ADO model discrimination favored the candidate model that was most similar to the data-generating hyperbolic model in predictions. In the second, the candidate models overlapped in predictions and thus the discriminability was low. The data-generating model, as a weighted mixture of power and exponential functions, behaved more “POW-like” when the weight on power function increased and more ”EXP-like” when the weight on power function decreased. We found that ADO model discrimination more strongly favored candidate model POW when the data-generating model behaved more “POW-like”, and vice versa. Finally, the so-called “similar in predictions”, or the model discrepancy was measured by utilizing the Kullback-Leibler divergence, which measures discrepancy between two probability distributions. In summary, in two simulations of model misspecification, ADO model discrimination succeeded in favoring the candidate model that was most similar to the data-generating model in their predictions.</p> 2012-12-17 English text The Ohio State University / OhioLINK http://rave.ohiolink.edu/etdc/view?acc_num=osu1347034580 http://rave.ohiolink.edu/etdc/view?acc_num=osu1347034580 unrestricted This thesis or dissertation is protected by copyright: all rights reserved. It may not be copied or redistributed beyond the terms of applicable copyright laws. |
collection |
NDLTD |
language |
English |
sources |
NDLTD |
topic |
Psychology |
spellingShingle |
Psychology Sun, Yinghao Adaptive Design Optimization for Model Discrimination under Model Misspecification |
author |
Sun, Yinghao |
author_facet |
Sun, Yinghao |
author_sort |
Sun, Yinghao |
title |
Adaptive Design Optimization for Model Discrimination under Model Misspecification |
title_short |
Adaptive Design Optimization for Model Discrimination under Model Misspecification |
title_full |
Adaptive Design Optimization for Model Discrimination under Model Misspecification |
title_fullStr |
Adaptive Design Optimization for Model Discrimination under Model Misspecification |
title_full_unstemmed |
Adaptive Design Optimization for Model Discrimination under Model Misspecification |
title_sort |
adaptive design optimization for model discrimination under model misspecification |
publisher |
The Ohio State University / OhioLINK |
publishDate |
2012 |
url |
http://rave.ohiolink.edu/etdc/view?acc_num=osu1347034580 |
work_keys_str_mv |
AT sunyinghao adaptivedesignoptimizationformodeldiscriminationundermodelmisspecification |
_version_ |
1719431048361672704 |