Reliability generalization meta-analysis of scales used to measure mathematics

The main purpose of this study is to obtain general reliability coefficients for common scales used to measure math anxiety. In addition, it is aimed to explore possible relationships between sample descriptors and mean reliability values, and to determine the proportions of studies that make induction through reporting or omission. For this purpose, nine different scales used by the researchers to measure math anxiety (Mathematics Anxiety Rating Scale [MARS], Revised Mathematics Anxiety Rating Scale [RMARS], Abbreviated Version of Mathematics Anxiety Rating Scale [MARS-AV], Mathematics Anxiety Rating Scale for Adolescents [MARSA], Suinn Mathematics Anxiety Rating Scale for Primary School Students [IMARS], Abbreviated Mathematics Anxiety Scale [AMARS], Revised Mathematics Anxiety Scale [RMAS], Suinn and Winston Mathematics Anxiety Rating Scale [SWMARS], Mathematics Anxiety Scale [MAS] were determined. General information and psychometric properties of these scales are presented under different headings. Published and unpublished studies reporting Cronbach's alpha coefficients since the scales were developed or revised were included in the analysis. The number of studies reporting alpha coefficients for the scales examined in this study ranged from 11 to 120. In order to obtain the average reliability coefficient according to the reliability generalization meta-analysis, the average reliability coefficient value of the studies transformed with the Bonett transformation method was analyzed in the metafor package in the R program by using the random effects model and the REML estimation method. The mean reliability coefficients of the scales examined in this study were estimated as follows: MARS=,96; RMARS=,93; MARS-AV=,93; MARSA=,91; IMARS=,89; AMARS=,86; RMAS=,90; SWMARS=,93; MAS=,89. In addition, heterogeneity and publication bias analyses were also performed. In order to investigate the source of heterogeneity, the effects of publication type, school level, country, continent, publication year, female participant ratio, number of items and sample size moderator variables on generalized alpha coefficients were investigated. In studies reporting alpha for all of the specified scales, the effect of the publication type variable on the average alpha coefficient was not found to be statistically significant. Five of the school level variable (MARS, RMARS, MARS-AV, AMARS, MAS), six of the country variable (MARS, RMARS, MARS-AV, MARSA, AMARS, MAS), four of the continent variable (MARS, RMARS, MARS-AV, MARSA), publication year variable for two (MARS, AMARS), the number of items for four (MARS, MARSA, IMARS, MAS), the female participant ratio for three (AMARS, RMAS, SWMARS) and the sample size variable for one scale (MAS) effects were found to be statistically significant. Based on the findings obtained as a result of the analyses, suggestions were made for the researchers and the limitations of the research were mentioned.