Normal view MARC view ISBD view

Determining Adequate Sample Sizes for Multiple Linear Regression used for Explanatory purposes / Rebernjak, Blaž.

By: Rebernjak, Blaž.
Material type: materialTypeLabelArticleDescription: 69-69.Other title: Determining Adequate Sample Sizes for Multiple Linear Regression used for Explanatory purposes [Naslov na engleskom:].Subject(s): 5.06 | Statistical power, regression analysis, Monte Carlo simulation, sample size hrv | Statistical power, regression analysis, Monte Carlo simulation, sample size engOnline resources: Elektronička verzija sažetka In: 19. Dani Ramira i Zorana Bujasa 22-25.04.2009. ; Zagreb, Hrvatska 19. Dani Ramira i Zorana Bujasa: program i sažeci priopćenja str. 69-69Ljubotina, Damir ; Kamenov, Željka ; Mikac, Una ; Urch, DraženSummary: There are numerous methods for determining minimum sample sizes when using multiple linear regression, from various rules of thumb to elaborate power analysis. Different methods, however, often yield surprisingly discrepant estimations. Monte Carlo simulation was used to examine models with varying number of predictor variables (2, 3, 4, 5, 7, 9, 11, 15), average intercorrelation between predictor variables (0, 0.1, 0.3, 0.5) and standardized regression coefficient of the specific predictor variable of interest (0.1, 0.3, 0.5), resulting in 96 different situations. Samples of varying sizes (from 25 to 5000) were drawn from multivariate normal distributions with 96 specified population correlation matrices. For each situation 5000 replications were made, resulting in a total of 7680000 analyses preformed. Based on the results for each situation, interpolation was used to approximate the sample size needed for the power to reach 0.80. Results show that all of the hypothesized variables influence minimum sample size with size of the regression coefficient being the strongest determinant. Limits of the present approach are discussed, and comparisons with proposed methods for determining adequate sample size are made for three hypothetical situations.
Tags from this library: No tags from this library for this title. Log in to add tags.
No physical items for this record

There are numerous methods for determining minimum sample sizes when using multiple linear regression, from various rules of thumb to elaborate power analysis. Different methods, however, often yield surprisingly discrepant estimations. Monte Carlo simulation was used to examine models with varying number of predictor variables (2, 3, 4, 5, 7, 9, 11, 15), average intercorrelation between predictor variables (0, 0.1, 0.3, 0.5) and standardized regression coefficient of the specific predictor variable of interest (0.1, 0.3, 0.5), resulting in 96 different situations. Samples of varying sizes (from 25 to 5000) were drawn from multivariate normal distributions with 96 specified population correlation matrices. For each situation 5000 replications were made, resulting in a total of 7680000 analyses preformed. Based on the results for each situation, interpolation was used to approximate the sample size needed for the power to reach 0.80. Results show that all of the hypothesized variables influence minimum sample size with size of the regression coefficient being the strongest determinant. Limits of the present approach are discussed, and comparisons with proposed methods for determining adequate sample size are made for three hypothetical situations.

Projekt MZOS 130-1301683-1402

ENG

There are no comments for this item.

Log in to your account to post a comment.

Powered by Koha