Assessing Heterogeneity in Meta-Analysis: Q Statistic or I

Department of Basic Psychology & Methodology, University of Murcia, Murcia, Spain.
Psychological Methods (Impact Factor: 4.45). 07/2006; 11(2):193-206. DOI: 10.1037/1082-989X.11.2.193
Source: PubMed


In meta-analysis, the usual way of assessing whether a set of single studies is homogeneous is by means of the Q test. However, the Q test only informs meta-analysts about the presence versus the absence of heterogeneity, but it does not report on the extent of such heterogeneity. Recently, the I(2) index has been proposed to quantify the degree of heterogeneity in a meta-analysis. In this article, the performances of the Q test and the confidence interval around the I(2) index are compared by means of a Monte Carlo simulation. The results show the utility of the I(2) index as a complement to the Q test, although it has the same problems of power with a small number of studies.

Download full-text


Available from: Fulgencio Marín-Martínez, Jun 06, 2014
529 Reads
  • Source
    • "rejected the null hypothesis of homogeneity, and í µí°¼ 2 indicated considerable nonrandom variability (Higgins & Thompson, 2002; Huedo-Medina et al., 2006) of 62.45% (95% CI [59.67, 64.95]). Specifically, an í µí°¼ 2 of 62.45% implies moderate to large nonrandom heterogeneity (Borenstein et al., 2009; Higgins & Thompson, 2002; Huedo-Medina et al., 2006), justifying moderator analyses. The distribution of these effect sizes can be found in the histogram in Figure 1. "
    [Show abstract] [Hide abstract]
    ABSTRACT: A meta-analysis assessed the behavioral impact of and psychological processes associated with presenting words connected to an action or a goal representation. The average and distribution of 352 effect sizes (analyzed using fixed-effects and random-effects models) was obtained from 133 studies (84 reports) in which word primes were incidentally presented to participants, with a non-opposite control group, prior to measuring a behavioral dependent variable. Findings revealed a small behavioral priming effect (𝑑𝐹𝐸 = 0.332, 𝑑𝑅𝐸 = 0.352), which was robust across methodological procedures and only minimally biased by the publication of positive (vs. negative) results. Theory testing analyses indicated that more valued behavior or goal concepts (e.g., associated with important outcomes or values) were associated with stronger priming effects than were less valued behaviors. Furthermore, there was some evidence of persistence of goal effects over time. These results support the notion that goal activation contributes over and above perception-behavior in explaining priming effects. In sum, theorizing about the role of value and satisfaction in goal activation pointed to stronger effects of a behavior or goal concept on overt action. There was no evidence that expectancy (ease of achieving the goal) moderated priming effects.
    Psychological Bulletin 08/2015; · 14.76 Impact Factor
  • Source
    • "Heterogeneity was estimated with the Cochran's Q-statistic, and p=0.05 was considered to be an indication of statistically significant heterogeneity. Also, effect of the heterogeneity quantified with the I 2 test (Huedo-Medina TB et al., 2006). As a guide, I 2 values ranged from 0 to 100%, and values of 25%, 50%, and 75% were considered to represent low, moderate, and high levels of heterogeneity, respectively. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Many studies have reported associations of the X-ray repair cross-complementing group 3 (XRCC3) Thr241Met polymorphism with colorectal cancer (CRC) risk, but the results remained controversial. Hence, we performed the present meta-analysis with different inheritance models. We searched the PubMed and Google scholar databases for studies relating to associations between XRCC3 Thr241Met polymorphism and risk of CRC. 16 studies with 5,193 cases and 6,645 controls were finally included into the meta-analysis. We found that the XRCC3 Thr241Met polymorphism was associated with increased CRC risk only under a dominant genetic model (CC+CT vs. TT: OR 0.575, 95%CI 0.498-1.665, p<0.001, Pheterogeneity =0.00, I2=83%). There was a significant association between XRCC3 Thr241Met polymorphism and CRC risk in Caucasian in the overall 8 studies under only in the heterozygote genetic model (CT vs. TT: OR=0.929, 95%CI =0.806-1.070, P=0.308, Pheterogeneity =0.002, I2=57%). Four studies evaluated the XRCC3 Thr241Met polymorphism and CRC risk in Asians. Two genetic models of the XRCC3 polymorphism were significantly correlated with increasing risk in Asians (dominant model: CC+CT vs. TT: OR= 0.609, 95%CI=411-0.902, P=0.013, Pheterogeneity =0.54, I2=0.00%; Allele model: C vs. T: OR=0.708, 95 %=CI 0.605-0.829, p=0.000, Pheterogeneity =0.000, I2=92%). The sensitivity analysis suggested stability of this meta-analysis and no publication bias was detected. In conclusion, this meta-analysis indicates that XRCC3 Thr241Met shows an increased CRC risk, particularly in Asians rather than Caucasians.
    Asian Pacific journal of cancer prevention: APJCP 03/2015; 16(6):2263-8. DOI:10.7314/APJCP.2015.16.6.2263 · 2.51 Impact Factor
  • Source
    • "Country Research design Components of school-based obesity prevention programs No. of children Mean age category of children (y) Program duration (y) Nutrition Physical activity Parental involvement Specialist involvement (e.g., nurse, nutritionist) Theory based Outcome measurement, BMI or skinfold thickness and positive (+) or negative (−) program effect Alexandrov [23] Moscow, USSR RCT 766 ≥10 ≥1 Yes Yes Yes No No BMI (+) Angelico [24] Italy Cohort 150 <10 ≥1 Yes Yes Yes No No BMI (−) Bayne- Smith [25] New York, USA RCT 442 ≥10 ≥1 Yes Yes No No No BMI (+) Bush [26] DC, USA RCT 431 ≥10 ≥1 Yes Yes Yes Yes Yes Skinfolds (−) Caballero [27] Arizona, New Mexico, South Dakota, USA RCT 1409 <10 ≥1 Yes Yes Yes No Yes BMI (+) Carrel [28] Wisconsin, USA RCT 50 ≥10 <1 Yes Yes No No No BMI (−) Donnelly [29] Nebraska, USA RCT 108 <10 ≥1 Yes Yes No Yes No BMI (+) Hansen [30] Denmark North Mix of case control and RCT 132 ≥10 <1 No Yes No No No BMI (−) Harrell [31] Carolina, USA RCT 1274 <10 <1 Yes Yes No Yes No BMI (−) Harrell [32] Mississippi, USA Intervention with control group 186 ≥10 <1 Yes Yes Yes Yes No BMI (−) James [33] England RCT 574 <10 ≥1 Yes No No No No BMI (+) James [34] England RCT 434 ≥10 ≥1 Yes No No No No BMI (+) Kain [35] Chile RCT 3086 <10 <1 Yes Yes Yes Yes No BMI Boys (+) BMI Girls (−) Killen [36] California, USA RCT 1130 ≥10 <1 Yes Yes No No Yes BMI (+) Kriemler [37] Switzerland RCT 502 <10 <1 No Yes No No No BMI (+) Luepker [38] California, Louisiana, Minnesota, Texas, USA RCT 3959 <10 ≥1 Yes Yes Yes No Yes BMI (−) Manios [39] Greece RCT 393 <10 ≥1 Yes Yes Yes No Yes BMI (+) Manios [40] Greece RCT 641 ≥10 ≥1 Yes Yes Yes Yes Yes BMI (+) Nader [41] USA RCT 3660 <10 ≥1 Yes Yes Yes No Yes BMI (+) Pangrazi [42] Arizona, USA RCT 599 <10 <1 No Yes No No No BMI (+) Puska [43] North Karelia, Finland RCT 851 <10 ≥1 Yes Yes Yes Yes No BMI (+) Ritenbaugh [44] New Mexico, USA Intervention with control group 102 ≥10 ≥1 Yes Yes No Yes No BMI (−) Robinson [45] California, USA RCT 192 ≥10 <1 No No Yes No Yes BMI (+) Sallis [46] California, USA RCT 955 <10 ≥1 No Yes No No No Skinfolds: Boys (+) for both groups Girls (+) for teacher-led Girls (−) for specialist-led Trevino [47] Texas, USA RCT 387 <10 <1 Yes Yes Yes Yes Yes BMI Boys (−) BMI Girls (+) Vandongen [48] Australia RCT 971 <10 <1 Yes Yes Yes Yes No BMI (+) for 6 groups and (−) for 4 groups Wardle [49] London, England Cohort 2730 <10 ≥1 No Yes No No No BMI (+) Fig. 2 "
    [Show abstract] [Hide abstract]
    ABSTRACT: Childhood obesity is a global concern. The objectives of this meta-analytical study were to evaluate the effectiveness of school-based childhood obesity prevention programs, and to examine program components (moderators). The methods included searching databases (PubMed, Google Scholar, and the university's EBSCOhost Web service) as well as handsearching reference lists of articles published in English. Selection criteria for studies to be included in the meta-analysis were limited to studies that reported body mass index (BMI) or skinfold thickness as outcome measures and were school-based obesity prevention interventions; cross-sectional design studies were excluded. We hypothesized the meta-analysis would yield a summary effect size of magnitude which would indicate that school-based interventions have been effective in improving children's BMI or skinfold thickness values. A total of 26 114 children from 27 school-based childhood obesity prevention programs provided 54 effect sizes. A random-effects model calculated a small summary effect size of 0.039 (95% confidence interval −0.013 to 0.092). Heterogeneity among studies was observed which disappeared after pooling studies that used a randomized controlled trial design with one program moderator (physical activity or nutrition). We failed to accept our hypothesis and concluded that overall, school-based interventions have not been effective for improving body mass index or skinfold thickness to curb childhood obesity; however, randomized controlled trials that focused on physical activity or nutrition appeared to produce promising results.
    Nutrition Research 01/2015; 35(3). DOI:10.1016/j.nutres.2015.01.002 · 2.47 Impact Factor
Show more