1987
DOI: 10.1037/0033-2909.101.1.126
|View full text |Cite
|
Sign up to set email alerts
|

Offending estimates in covariance structure analysis: Comments on the causes of and solutions to Heywood cases.

Abstract: In this article we discuss, illustrate, and compare the relative efficacy of three recommended approaches for handling negative error variance estimates (i.e., Heywood cases): (a) setting the offending estimate to zero, (b) adopting a model parameterization that ensures positive error variance estimates, and (c) using models with equality constraints that ensure nonnegative (but possibly zero) error variance estimates. The three approaches are evaluated in two distinct situations: Heywood cases caused by lack … Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

1
179
0
9

Year Published

1995
1995
2021
2021

Publication Types

Select...
8
1

Relationship

0
9

Authors

Journals

citations
Cited by 285 publications
(191 citation statements)
references
References 29 publications
1
179
0
9
Order By: Relevance
“…Researchers have recommended sample sizes larger than 200 for greater validity of structural modeling 58,59 and argued that insufficient data is a latent cause for Heywood cases in which standardized loading is larger than one and error variance is negative. 60,61 However, both the measurement and structural models yielded no such cases, and our sample size is arguably acceptable given the experimental nature of this study, which offers the ability to isolate the attribution of observed effects solely to the variation in screen size. Second, controlling for individual differences, such as gender, age, and product knowledge, could have increased the exploratory strength of the study.…”
Section: Effects Of Screen Size On Smartphone Adoption 469mentioning
confidence: 99%
“…Researchers have recommended sample sizes larger than 200 for greater validity of structural modeling 58,59 and argued that insufficient data is a latent cause for Heywood cases in which standardized loading is larger than one and error variance is negative. 60,61 However, both the measurement and structural models yielded no such cases, and our sample size is arguably acceptable given the experimental nature of this study, which offers the ability to isolate the attribution of observed effects solely to the variation in screen size. Second, controlling for individual differences, such as gender, age, and product knowledge, could have increased the exploratory strength of the study.…”
Section: Effects Of Screen Size On Smartphone Adoption 469mentioning
confidence: 99%
“…In this model the slope variance of younger adults did not differ reliably from zero. Hence the slope variance in younger adults and the slopeintercept covariance in both age groups were fixed to zero (Dillon, Kumar, & Mulani, 1987;Ghisletta & Lindenberger, 2005). The resulting model showed acceptable fit to the data, x 2 011.59, df 09, CFI 0.98 and RMSEA 00.10 [90% CI .00 Á .24] and did not differ from the initial unrestricted model, Dx 2 01.64, df 03, p .05.…”
Section: Age Differences In Hits and Famentioning
confidence: 99%
“…A Heywood case was identified (item 5: standardized loading = 1.05) when running the three-factor model. To correct for this, the loading was fixed to unity (Dillon, Kumar, & Mulani, 1987), but model fit statistics, loadings and error terms did not alter significantly indicating that the Heywood case is most likely caused by sampling fluctuations. The standardized factor loadings (McDonald & Ringo Ho, 2002) for the threefactor model are given in Table 2 (loadings for the four-factor model can be obtained from the first author).…”
Section: Confirmatory Factor Analysismentioning
confidence: 99%