東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

Properties of Partially Convergent M...

Dogucu, Mine.

FindBook

Google Book

Amazon

博客來

Properties of Partially Convergent Models and Effect of Re-Imputation on These Properties.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Properties of Partially Convergent Models and Effect of Re-Imputation on These Properties./
作者:	Dogucu, Mine.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2017,
面頁冊數:	138 p.
附註:	Source: Dissertations Abstracts International, Volume: 79-08, Section: B.
Contained By:	Dissertations Abstracts International79-08B.
標題:	Educational tests & measurements. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10702641
ISBN:	9780355442229

Properties of Partially Convergent Models and Effect of Re-Imputation on These Properties.
Dogucu, Mine.

Properties of Partially Convergent Models and Effect of Re-Imputation on These Properties. - Ann Arbor : ProQuest Dissertations & Theses, 2017 - 138 p.

Source: Dissertations Abstracts International, Volume: 79-08, Section: B.

Thesis (Ph.D.)--The Ohio State University, 2017.

This item must not be sold to any third party vendors.

When researchers fit statistical models to multiply imputed datasets, they have to fit the model separately for each imputed dataset. Since there are multiple datasets, there are always multiple sets of model results. It is possible for some of these sets of results not to converge while some do converge. This study examined occurrence of such a problem - partial convergence problem and inspected four outcomes of partially convergent models: proportion of convergence, percent parameter bias, root mean square error, and coverage rate of the true population parameter in the confidence interval of the parameter estimates. In this Monte Carlo simulation study complete data with standard normal distribution were simulated based on a single CFA model with 3 factors each with 3 indicators. True factor loadings for each factor varied as .4, .6, and .8. A single factor covariance of .4 was set as the true population parameter. In addition to variation of the factor loadings, sample size levels varied as 60, 80, 100, 150, 200, 300, and 400 and rate of missingness varied as 5%, 10%, 25%, and 50%. Partial convergence problem was not observed for sample size levels of 300 and 400. Thus, these conditions were not examined any further. It was examined that partial convergence problem was more likely to occur when the sample size was small and rate of missingness was high. It was observed that factor loading estimations were unbiased, however, there were two cells (sample of 60 with 25% missingness and sample size of 150 with 50% missingness) where factor covariance estimations were biased. Even though, overall for most cells parameter estimates were unbiased, root mean square errors were high across sample sizes for the 50% missingness condition which indicates that the parameter estimates for the 50% missingness were inconsistent across replications. Almost for all conditions, confidence intervals of the parameter estimates did not contain the true population parameter at acceptable rate - 90% or higher. A different solution to the partial convergence problem was examined in this study. For the sets of results that did not converge, additional imputations were conducted to examine whether any of the outcomes would be improved. The only improvement was observed in the proportion of convergence outcome, indicating, that re-imputation may result in fully convergence of models, however, this will have no effect on percent parameter bias, root mean square error or the coverage rate of the confidence intervals. Recommendations for researchers who deal with the partial convergence problem are provided. Limitations of the study are acknowledged and suggestions for future research are given.

ISBN: 9780355442229Subjects--Topical Terms:

3168483
Educational tests & measurements.
Subjects--Index Terms:

Missing data

Properties of Partially Convergent Models and Effect of Re-Imputation on These Properties.
LDR:04028nmm a2200385 4500 001 2269013
005 20200908082311.5
008 220629s2017 ||||||||||||||||| ||eng d
020 $a 9780355442229
035 $a (MiAaPQ)AAI10702641
035 $a (MiAaPQ)OhioLINK:osu1499600301334388
035 $a AAI10702641
040 $a MiAaPQ $c MiAaPQ
100 1 $a Dogucu, Mine. $3 3546317
245 1 0 $a Properties of Partially Convergent Models and Effect of Re-Imputation on These Properties.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2017
300 $a 138 p.
500 $a Source: Dissertations Abstracts International, Volume: 79-08, Section: B.
500 $a Publisher info.: Dissertation/Thesis.
500 $a Advisor: Lomax, Richard.
502 $a Thesis (Ph.D.)--The Ohio State University, 2017.
506 $a This item must not be sold to any third party vendors.
506 $a This item must not be added to any third party search indexes.
520 $a When researchers fit statistical models to multiply imputed datasets, they have to fit the model separately for each imputed dataset. Since there are multiple datasets, there are always multiple sets of model results. It is possible for some of these sets of results not to converge while some do converge. This study examined occurrence of such a problem - partial convergence problem and inspected four outcomes of partially convergent models: proportion of convergence, percent parameter bias, root mean square error, and coverage rate of the true population parameter in the confidence interval of the parameter estimates. In this Monte Carlo simulation study complete data with standard normal distribution were simulated based on a single CFA model with 3 factors each with 3 indicators. True factor loadings for each factor varied as .4, .6, and .8. A single factor covariance of .4 was set as the true population parameter. In addition to variation of the factor loadings, sample size levels varied as 60, 80, 100, 150, 200, 300, and 400 and rate of missingness varied as 5%, 10%, 25%, and 50%. Partial convergence problem was not observed for sample size levels of 300 and 400. Thus, these conditions were not examined any further. It was examined that partial convergence problem was more likely to occur when the sample size was small and rate of missingness was high. It was observed that factor loading estimations were unbiased, however, there were two cells (sample of 60 with 25% missingness and sample size of 150 with 50% missingness) where factor covariance estimations were biased. Even though, overall for most cells parameter estimates were unbiased, root mean square errors were high across sample sizes for the 50% missingness condition which indicates that the parameter estimates for the 50% missingness were inconsistent across replications. Almost for all conditions, confidence intervals of the parameter estimates did not contain the true population parameter at acceptable rate - 90% or higher. A different solution to the partial convergence problem was examined in this study. For the sets of results that did not converge, additional imputations were conducted to examine whether any of the outcomes would be improved. The only improvement was observed in the proportion of convergence outcome, indicating, that re-imputation may result in fully convergence of models, however, this will have no effect on percent parameter bias, root mean square error or the coverage rate of the confidence intervals. Recommendations for researchers who deal with the partial convergence problem are provided. Limitations of the study are acknowledged and suggestions for future research are given.
590 $a School code: 0168.
650 4 $a Educational tests & measurements. $3 3168483
650 4 $a Statistics. $3 517247
650 4 $a Quantitative psychology. $3 2144748
653 $a Missing data
653 $a Model convergence
653 $a Multiple imputation
690 $a 0288
690 $a 0463
690 $a 0632
710 2 $a The Ohio State University. $b Educational Studies. $3 3173395
773 0 $t Dissertations Abstracts International $g 79-08B.
790 $a 0168
791 $a Ph.D.
792 $a 2017
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=10702641