Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Quantifying Selective Reporting and the Proteus Phenomenon for Multiple Datasets with Similar Bias

Figure 2

Results for the high-resolution models and the one category model (Model 1).

Estimates for the high-resolution random-effects model are shown in solid lines, estimates for the fixed-effect model in dashed lines. Error bars show standard errors of the estimate. (A) High resolution model. For both the fixed-effect and random-effects model, the weights drop rapidly as |Z| decreases. At about |Z| = 1.64, the weights hit a bottom and remain relatively constant in the central intervals. This is what one would expect for publication bias: Non-significant results are subject to a similar bias, irrespective of the Z-value. (B) Model 1. In the random-effects model, bias in the mid interval is estimated as w(S) ≈ 0.7. Under the fixed-effect model, we estimate w(S) ≈ 0.4. The estimates for the bias in the fixed-effect model tend to be higher than in the random-effects model, because in the random-effects model, a high between-study variance offers an additional explanation for an excess of formally significant results.

Figure 2

doi: https://doi.org/10.1371/journal.pone.0018362.g002