2001
DOI: 10.1177/109442810141003
|View full text |Cite
|
Sign up to set email alerts
|

Correcting the Effect Size of d for Range Restriction and Unreliability

Abstract: Pearson product moment correlations are often corrected for statistical artifacts such as range restriction and unreliability. Formulas have long existed to make such corrections. However, other effect size estimates are rarely corrected for these artifacts, in spite of the fact that there is an established mathematical link between the correlation and some effect size estimates. Correlations and other effect sizes are therefore vulnerable to the same artifacts. The authors take a common effect size estimate, … Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1
1

Citation Types

1
91
0
1

Year Published

2007
2007
2022
2022

Publication Types

Select...
7
2

Relationship

0
9

Authors

Journals

citations
Cited by 83 publications
(93 citation statements)
references
References 34 publications
1
91
0
1
Order By: Relevance
“…On a similar note, a common practice within our sample of studies was to split the items on Gf outcome measures in half in order to have a comparable pretest and posttest version. While this is effective in reducing test-retest practice effects, an unfortunate consequence is a reduction in measurement reliability , which also causes downward biases in ES due to an increase in error variance that weakens the strength of its correlations (Bobko et al, 2001). Taken together, we expect that the results reported in this meta-analysis represent a low-end estimate of the true extent of improvement that n-back training can have on measures of Gf.…”
Section: Discussionmentioning
confidence: 71%
See 1 more Smart Citation
“…On a similar note, a common practice within our sample of studies was to split the items on Gf outcome measures in half in order to have a comparable pretest and posttest version. While this is effective in reducing test-retest practice effects, an unfortunate consequence is a reduction in measurement reliability , which also causes downward biases in ES due to an increase in error variance that weakens the strength of its correlations (Bobko et al, 2001). Taken together, we expect that the results reported in this meta-analysis represent a low-end estimate of the true extent of improvement that n-back training can have on measures of Gf.…”
Section: Discussionmentioning
confidence: 71%
“…Within this restricted range of young adults, the majority of participants were college undergraduates, thereby skewing our samples even younger (mean age ± SD: 22.85 ± 2.60). A common property of statistics holds that sampling from restricted ranges of the total population usually biases ES downward, due to reduced variability (Bobko, Roth, & Bobko, 2001;Fritz, Morris, & Richler, 2012). On a similar note, a common practice within our sample of studies was to split the items on Gf outcome measures in half in order to have a comparable pretest and posttest version.…”
Section: Discussionmentioning
confidence: 99%
“…Although techniques exist to attempt to correct for unreliability (e.g., Hunter & Schmidt, 2004;Schmidt et al, 2009), we chose to allow the two potential opposing biases to help balance each other, with the aim of maximizing the accuracy of our estimates. Thus, in our analysis, the potential publication bias indicated by the funnel plot (which would lead to an overestimation of the effect size; Duval & Tweedie, 2000) and the unreliability of measures we know to be present in the sample studies (which leads to an underestimation of the effect size; Bobko, Roth, & Bobko, 2001, Card, 2012Hunter & Schmidt, 2004) are pushing the estimated average effect size in opposite directions. Though it is impossible to know which of these two biases is bigger, or whether they are of similar size to perfectly cancel each other out, they should counteract each other to some extent in terms of the mean effect size estimate.…”
Section: Publication Biasmentioning
confidence: 99%
“…Although the version SILL 7 includes 50 questions and 6 subscales, in the present study due to some infeasibilities 'only 35 questions and 3 subscales as metacognitive (items 15-23), cognitive(items [1][2][3][4][5][6][7][8][9][10][11][12][13][14], and social/ affective strategies(items 24-35) were included.…”
Section: Participantsmentioning
confidence: 99%