Skip to contents

Perform a p-curve analysis

Source: R/dmetar_contributions.R
pcurve.Rd

This function performs a \(p\)-curve analysis using a meta object or calculated effect size data. See the dmetar documentation for examples.

Usage

pcurve(x, effect.estimation = FALSE, N, dmin = 0, dmax = 1)

Arguments

x

Either an object of class meta, generated by the metagen, metacont, metacor, metainc, or metabin function, or a dataframe containing the calculated effect size (named TE, log-transformed if based on a ratio), standard error (named seTE) and study label (named studlab) for each study.

effect.estimation

Logical. Should the true effect size underlying the p-curve be estimated? If set to TRUE, a vector containing the total sample size for each study must be provided for N. FALSE by default.

N

A numeric vector of same length as the number of effect sizes included in x specifying the total sample size \(N\) corresponding to each effect. Only needed if effect.estimation = TRUE.

dmin

If effect.estimation = TRUE: lower limit for the effect size (\(d\)) space in which the true effect size should be searched. Must be greater or equal to 0. Default is 0.

dmax

If effect.estimation = TRUE: upper limit for the effect size (\(d\)) space in which the true effect size should be searched. Must be greater than 0. Default is 1.

Value

Returns a plot and main results of the pcurve analysis:

  • P-curve plot: A plot displaying the observed \(p\)-curve and significance results for the right-skewness and flatness test.

  • Number of studies: The number of studies provided for the analysis, the number of significant \(p\)-values included in the analysis, and the number of studies with \(p<0.025\) used for the half-curve tests.

  • Test results: The results for the right-skewness and flatness test, including the \(p_{binomial}\) value, as well as the \(z\) and \(p\) value for the full and half-curve test.

  • Power Estimate: The power estimate and 95% confidence interval.

  • Evidential value: Two lines displaying if evidential value is present and/or absent/inadequate based on the results (using the guidelines by Simonsohn et al., 2015, see Details).

  • True effect estimate: If effect.estimation is set to TRUE, the estimated true effect \(\hat{d}\) is returned additionally.

If results are saved to a variable, a list of class pcurve containing the following objects is returned:

  • pcurveResults: A data frame containing the results for the right-skewness and flatness test, including the \(p_{binomial}\) value, as well as the \(z\) and \(p\) value for the full and half-curve test.

  • Power: The power estimate and 95% confidence interval.

  • PlotData: A data frame with the data used in the \(p\)-curve plot.

  • Input: A data frame containing the provided effect sizes, calculated \(p\)-values and individual results for each included (significant) effect.

  • EvidencePresent, EvidenceAbsent, kInput, kAnalyzed, kp0.25: Further results of the \(p\)-curve analysis, including the presence/absence of evidence interpretation, and number of provided/significant/\(p<0.025\) studies.

  • I2: \(I^2\)-Heterogeneity of the studies provided as input (only when x is of class meta).

  • class.meta.object: class of the original object provided in x.

Details

P-curve Analysis

\(P\)-curve analysis (Simonsohn, Simmons & Nelson, 2014, 2015) has been proposed as a method to detect \(p\)-hacking and publication bias in meta-analyses.

\(P\)-Curve assumes that publication bias is not only generated because researchers do not publish non-significant results, but also because analysts “play” around with their data ("\(p\)-hacking"; e.g., selectively removing outliers, choosing different outcomes, controlling for different variables) until a non-significant finding becomes significant (i.e., \(p<0.05\)).

The method assumes that for a specific research question, \(p\)-values smaller 0.05 of included studies should follow a right-skewed distribution if a true effect exists, even when the power in single studies was (relatively) low. Conversely, a left-skewed \(p\)-value distribution indicates the presence of \(p\)-hacking and absence of a true underlying effect. To control for "ambitious" \(p\)-hacking, \(P\)-curve also incorporates a "half-curve" test (Simonsohn, Simmons & Nelson, 2014, 2015).

Simonsohn et al. (2014) stress that \(p\)-curve analysis should only be used for test statistics which were actually of interest in the context of the included study, and that a detailed table documenting the reported results used in for the \(p\)-curve analysis should be created before communicating results (link).

Implementation in the function

To generate the \(p\)-curve and conduct the analysis, this function reuses parts of the R code underlying the P-curve App 4.052 (Simonsohn, 2017). The effect sizes included in the meta object or data.frame provided for x are transformed into \(z\)-values internally, which are then used to calculate p-values and conduct the Stouffer and Binomial test used for the \(p\)-curve analysis. Interpretations of the function concerning the presence or absence/inadequateness of evidential value are made according to the guidelines described by Simonsohn, Simmons and Nelson (2015):

  • Evidential value present: The right-skewness test is significant for the half curve with \(p<0.05\) or the \(p\)-value of the right-skewness test is \(<0.1\) for both the half and full curve.

  • Evidential value absent or inadequate: The flatness test is \(p<0.05\) for the full curve or the flatness test for the half curve and the binomial test are \(p<0.1\).

For effect size estimation, the pcurve function implements parts of the loss function presented in Simonsohn, Simmons and Nelson (2014b). The function generates a loss function for candidate effect sizes \(\hat{d}\), using \(D\)-values in a Kolmogorov-Smirnov test as the metric of fit, and the value of \(\hat{d}\) which minimizes \(D\) as the estimated true effect.

It is of note that a lack of robustness of \(p\)-curve analysis results has been noted for meta-analyses with substantial heterogeneity (van Aert, Wicherts, & van Assen, 2016). Following van Aert et al., adjusted effect size estimates should only be reported and interpreted for analyses with \(I^2\) values below 50 percent. A warning message is therefore printed by the pcurve function when x is of class meta and the between-study heterogeneity of the meta-analysis is substantial (i.e., \(I^2\) greater than 50 percent).

References

Harrer, M., Cuijpers, P., Furukawa, T.A, & Ebert, D. D. (2019). Doing Meta-Analysis in R: A Hands-on Guide. DOI: 10.5281/zenodo.2551803. Chapter 9.2.

Simonsohn, U., Nelson, L. D., & Simmons, J. P. (2014a). P-curve: a Key to the File-drawer. Journal of Experimental Psychology, 143(2), 534.

Simonsohn, U., Nelson, L. D. & Simmons, J. P. (2014b). P-Curve and Effect Size: Correcting for Publication Bias Using Only Significant Results. Perspectives on Psychological Science 9(6), 666–81.

Simonsohn, U., Nelson, L. D. & Simmons, J. P. (2015). Better P-Curves: Making P-Curve Analysis More Robust to Errors, Fraud, and Ambitious P-Hacking, a Reply to Ulrich and Miller (2015). Journal of Experimental Psychology, 144(6), 1146-1152.

Simonsohn, U. (2017). R code for the P-Curve App 4.052. http://p-curve.com/app4/pcurve_app4.052.r (Accessed 2019-08-16).

Van Aert, R. C., Wicherts, J. M., & van Assen, M. A. (2016). Conducting meta-analyses based on p values: Reservations and recommendations for applying p-uniform and p-curve. Perspectives on Psychological Science, 11(5), 713-729.

Author

Mathias Harrer & David Daniel Ebert