2011
DOI: 10.1002/sim.4451
|View full text |Cite
|
Sign up to set email alerts
|

Higher‐order likelihood inference in meta‐analysis and meta‐regression

Abstract: This paper investigates the use of likelihood methods for meta-analysis, within the random-effects models framework. We show that likelihood inference relying on first-order approximations, while improving common meta-analysis techniques, can be prone to misleading results. This drawback is very evident in the case of small sample sizes, which are typical in meta-analysis. We alleviate the problem by exploiting the theory of higher-order asymptotics. In particular, we focus on a second-order adjustment to the … Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

1
86
0

Year Published

2013
2013
2022
2022

Publication Types

Select...
8

Relationship

1
7

Authors

Journals

citations
Cited by 45 publications
(87 citation statements)
references
References 39 publications
(52 reference statements)
1
86
0
Order By: Relevance
“…However, the PE method can only be performed when the number of studies is larger than five, whereas many meta-analyses are smaller [15]. Several other methods have been developed, like the Quantile Approximation (QA) method [28], the Profile Likelihood approach [29], natural weighting instead of empirically based weighting of studies [30], use of fixed effects estimates with a random effects approach to heterogeneity [31] and more recently, higher-order likelihood inference methods [32]. However, most of these methods are based on asymptotic statistics and they may therefore be less robust in case of a limited number of trials, or they remain difficult to use in practice, because no statistical packages are available to perform them and it is very difficult to carry out the calculations with standard software.…”
Section: Discussionmentioning
confidence: 99%
“…However, the PE method can only be performed when the number of studies is larger than five, whereas many meta-analyses are smaller [15]. Several other methods have been developed, like the Quantile Approximation (QA) method [28], the Profile Likelihood approach [29], natural weighting instead of empirically based weighting of studies [30], use of fixed effects estimates with a random effects approach to heterogeneity [31] and more recently, higher-order likelihood inference methods [32]. However, most of these methods are based on asymptotic statistics and they may therefore be less robust in case of a limited number of trials, or they remain difficult to use in practice, because no statistical packages are available to perform them and it is very difficult to carry out the calculations with standard software.…”
Section: Discussionmentioning
confidence: 99%
“…For example, empirical coverages of confidence intervals are lower than the nominal level, and hypothesis tests can result in erroneous conclusions. In meta-analysis, recent works investigate the inaccuracy of first-order likelihood solutions when the sample size is small 10,18 and when the sample size within each study included in the meta-analysis is small as well. 19 When the reduced sample size cannot guarantee accuracy of asymptotic normality, the routine use of r P is discouraged and alternative solutions have been developed.…”
Section: First-order and Higher-order Likelihood Inferencementioning
confidence: 99%
“…For the more general case of meta-analysis and meta-regression, results in Guolo 10 highlight the advantages of using r P in place of r P in terms of accuracy of inferential conclusions for small to moderate sample sizes.…”
Section: Examplementioning
confidence: 99%
See 1 more Smart Citation
“…68 69 Therefore, to pool outcome data for trials that make direct comparisons between interventions and alternatives, we will use the likelihood profile approach. 70 We will pool cross-over trials with parallel design RCTs using methods outlined in the Cochrane handbook to derive effect estimates. 66 Specifically, we will perform a paired t test for each crossover trial if any of the following are available: (1) the individual participant data; (2) the mean and SD or SE of the participant-specific differences, and between the intervention and control measurement; (3) the mean difference (MD) and one of the following: (a) a t-statistic from a paired t test; (b) a p value from a paired t-test; (c) a CI from a paired analysis; or (4) a graph of measurements of the intervention arm and control arm from which we can extract individual data values, so long as the matched measurement for each individual can be identified.…”
Section: Direct Comparisons Meta-analysesmentioning
confidence: 99%