Category Archives: Zcurve

An Introduction to Z-Curve: A method for estimating mean power after selection for significance (replicability)

Observed Power, Post-Hoc Power, Posteriori Power Analysis, Power, Replicability, Z-Curve, Zcurve

Since 2015, Jerry Brunner and I have been working on a statistical tool that can estimate mean (statitical) power for a set of studies with heterogeneous sample sizes and effect sizes (heterogeneity in non-centrality parameters and true power).   This method corrects for the inflation in mean observed power that is introduced by the selection for statistical significance.   Knowledge about mean power makes it possible to predict the success rate of exact replication studies.   For example, if a set of studies with mean power of 60% were replicated exactly (including sample sizes), we would expect that 60% of the replication studies produce a significant result again.

Our latest manuscript is a revision of an earlier manuscript that received a revise and resubmit decision from the free, open-peer-review journal Meta-Psychology.  We consider it the most authoritative introduction to z-curve that should be used to learn about z-curve, critic z-curve, or as a citation for studies that use z-curve.

Cite as “submitted for publication”.

Final.Revision.874-Manuscript in PDF-2236-1-4-20180425 mva final (002)

Feel free to ask questions, provide comments, and critic our manuscript in the comments section.  We are proud to be an open science lab, and consider criticism an opportunity to improve z-curve and our understanding of power estimation.

R-CODE
Latest R-Code to run Z.Curve (Z.Curve.Public.18.10.28).
[updated 18/11/17]   [35 lines of code]
call function  mean.power = zcurve(pvalues,Plot=FALSE,alpha=.05,bw=.05)[1]

Z-Curve related Talks
Presentation on Z-curve and application to BS Experimental Social Psychology and (Mostly) WS-Cognitive Psychology at U Waterloo (November 2, 2018)
[Powerpoint Slides]

 

 

 

 

Can the Bayesian Mixture Model Estimate the Percentage of False Positive Results in Psychology Journals?

Bayesian Statistics, Mixture Model, Z-Curve, Zcurve

A method revolution is underway in psychological science.  In 2011, an article published in JPSP-ASC made it clear that experimental social psychologists were publishing misleading p-values because researchers violated basic principles of significance testing  (Schimmack, 2012; Wagenmakers et al., 2011).  Deceptive reporting practices led to the publication of mostly significant results, while many non-significant results were not reported.  This selective publishing of results dramatically increases the risk of a false positive result from the nominal level of 5% that is typically claimed in publications that report significance tests  (Sterling, 1959).

Although experimental social psychologists think that these practices are defensible, no statistician would agree with them.  In fact, Sterling (1959) already pointed out that the success rate in psychology journals is too high and claims about statistical significance are meaningless.  Similar concerns were raised again within psychology (Rosenthal, 1979), but deceptive practices remain acceptable until today (Kitayama, 2018). As a result, most published results in social psychology do not replicate and cannot be trusted (Open Science Collaboration, 2015).

For non-methodologists it can be confusing to make sense of the flood of method papers that have been published in the past years.  It is therefore helpful to provide a quick overview of methodological contributions concerned with detection and correction of biases.

First, some methods focus on effect sizes, (pcurve2.0; puniform), whereas others focus on strength of evidence (Test of Excessive Significance; Incredibility Index; R-Index, Pcurve2.1; Pcurve4.06; Zcurve).

Another important distinction is between methods that assume a fixed parameter and methods that assume heterogeneity.   If all studies have a common effect size or the same strength of evidence,  it is relatively easy to demonstrate bias and to correct for bias (Pcurve2.1; Puniform; TES).  However, heterogeneity in effect sizes or sampling error produces challenges.  Relatively few methods have been developed for this challenging, yet realistic scenario.  For example, Ioannidis and Trikalonis (2005) developed a method to reveal publication bias that assumes a fixed effect size across studies, while allowing for variation in sampling error, but this method can be biased if there is heterogeneity in effect sizes.  In contrast, I developed the Incredibilty-Index (also called Magic Index) to allow for heterogeneity in effect sizes and sampling error (Schimmack, 2012).

Following my work on bias detection in heterogeneous sets of studies, I started working with Jerry Brunner on methods that can estimate average power of a heterogeneous set of studies that are selected for significance.  I first published this method on my blog in June 2015, when I called it post-hoc power curves.   These days, the term Zcurve is used more often to refer to this method.  I illustrated the usefulness of Zcurve in various posts in the Psychological Methods Discussion Group.

In September, 2015 I posted replicability rankings of social psychology departments using this method. the post generated a lot of discussions and a question about the method.  Although the details were still unpublished, I described the main approach of the method.  To deal with heterogeneity, the method uses a mixture model.

EJ.Mixture.png

In 2016, Jerry Brunner and I submitted a manuscript for publication that compared four methods for estimating average power of heterogeneous studies selected for significance (Puniform1.1; Pcurve2.1; Zcurve & a Maximul Likelihood Method).  In this article, the mixture model, Zcurve, outperformed other methods, including a maximum-likelihood method developed by Jerry Brunner. The manuscript was rejected from Psychological Methods.

In 2017, Gronau, Duizer, Bakker, and Eric-Jan Wagenmakers published an article titled “A Bayesian Mixture Modeling of Significant p Values: A Meta-Analytic Method to Estimate the Degree of Contamination From H0”  in the Journal of Experimental Psychology: General.  The article did not mention z-curve, presumably because it was not published in a peer-reviewed journal.

Although a reference to our mixture model would have been nice, the Bayesian Mixture Model differs in several ways from Zcurve.  This blog post examines the similarities and differences between the two mixture models, it shows that BMM fails to provide useful estimates with simulations and social priming studies, and it explains why BMM fails. It also shows that Zcurve can provide useful information about replicability of social priming studies, while the BMM estimates are uninformative.

Aims

The Bayesian Mixture Model (BMM) and Zcurve have different aims.  BMM aims to estimate the percentage of false positives (significant results with an effect size of zero). This percentage is also called the False Discovery Rate (FDR).

FDR = False Positives / (False Positives + True Positives)

Zcurve aims to estimate the average power of studies selected for significance. Importantly, Brunner and Schimmack use the term power to refer to the unconditional probability of obtaining a significant result and not the common meaning of power as being conditional on the null-hypothesis being false. As a result, Zcurve does not distinguish between false positives with a 5% probability of producing a significant result (when alpha = .05) and true positives with an average probability between 5% and 100% of producing a significant result.

Average unconditional power is simply the percentage of false positives times alpha plus the average conditional power of true positive results (Sterling et al., 1995).

Unconditional Power = False Positives * Alpha + True Positives * Mean(1 – Beta)

Zcurve therefore avoids the thorny issue of defining false positives and trying to distinguish between false positives and true positives with very small effect sizes and low power.

Approach 

BMM and zcurve use p-values as input.  That is, they ignore the actual sampling distribution that was used to test statistical significance.  The only information that is used is the strength of evidence against the null-hypothesis; that is, how small the p-value actually is.

The problem with p-values is that they have a specified sampling distribution only when the null-hypothesis is true. When the null-hypothesis is true, p-values have a uniform sampling distribution.  However, this is not useful for a mixture model, because a mixture model assumes that the null-hypothesis is sometimes false and the sampling distribution for true positives is not defined.

Zcurve solves this problem by using the inverse normal distribution to convert all p-values into absolute z-scores (abs(z) = -qnorm(p/2).  Absolute z-scores are used because F-tests or two-sided t-tests do not have a sign and a test score of 0 corresponds to a probability of 1.  Thus, the results do not say anything about the direction of an effect, while the size of the p-value provides information about the strength of evidence.

BMM also transforms p-values. The only difference is that BMM uses the full normal distribution with positive and negative z-scores  (z = qnorm(p)). That is, a p-value of .5 corresponds to a z-score of zero; p-values greater than .5 would be positive, and p-values less than .5 are assigned negative z-scores.  However, because only significant p-values are selected, all z-scores are negative in the range from -1.65 (p = .05, one-tailed) to negative infinity (p = 0).

The non-centrality parameter (i.e., the true parameter that generates the sampling dstribution) is simply the mean of the normal distribution. For the null-hypothesis and false positives, the mean is zero.

Zcurve and BMM differ in the modeling of studies with true positive results that are heterogeneous.  Zcurve uses several normal distributions with a standard deviation of 1 that reflects sampling error for z-tests.  Heterogeneity in power is modeled by varying means of normal distributions, where power increases with increasing means.

BMM uses a single normal distribution with varying standard deviation.  A wider distribution is needed to predict large observed z-scores.

The main difference between Zcurve and BMM is that Zcurve either does not have fixed means (Brunner & Schimmack, 2016) or has fixed means, but does not interpret the weight assigned to a mean of zero as an estimate of false positives (Schimmack & Brunner, 2018).  The reason is that the weights attached to individual components are not very reliable estimates of the weights in the data-generating model.  Importantly, this is not relevant for the goal of zurve to estimate average power because the weighted average of the components of the model is a good estimate of the average true power in the data-generating model, even if the weights do not match the weights of the data-generating model.

For example, Zcurve does not care whether 50% average power is produced by a mixture of 50% false positives and 50% true positives with 95% power or 50% of studies with 20% power and 50% studies with 80% power. If all of these studies were exactly replicated, they are expected to produce 50% significant results.

BMM uses the weights assigned to the standard normal with a mean of zero as an estimate of the percentage of false positive results.  It does not estimate the average power of true positives or average unconditional power.

Given my simulation studies with zcruve, I was surprised that BBM solved a problem that weights of individual components cannot be reliably estimated because the same distribution of p-values can be produced by many mixture models with different weights.  The next section examines how BMM tries to estimate the percentage of false positives from the distribution of p-values.

A Bayesian Approach

Another difference between BMM and Zcurve is that BMM uses prior distributions, whereas Zcurve does not.  Whereas Zcurve makes no assumptions about the percentage of false positives, BMM uses a uniform distribution with values from 0 to 1 (100%) as a prior.  That is, it is equally likely that the percentage of false positives is 0%, 100%, or any value in between.  A uniform prior is typically justified as being agnostic; that is, no subjective assumptions bias the final estimate.

For the mean of the true positives, the authors use a truncated normal prior, which they also describe as a folded standard normal.  They justify this prior as reasonable based on extensive simulation studies.

Most important, however, is the parameter for the standard deviation.  The prior for this parameter was a uniform distribution with values between 0 and 1.   The authors argue that larger values would produce too many p-values close to 1.

“implausible prediction that p values near 1 are more common under H1 than under H0” (p 1226). 

But why would this be implausible.  If there are very few false positives and many true positives with low power, most p-values close to 1 would be the result of  true positives (H1) than of false positives (H0).

Thus, one way BMM is able to estimate the false discovery rate is by setting the standard deviation in a way that there is a limit to the number of low z-scores that are predicted by true positives (H1).

Although understanding priors and how they influence results is crucial for meaningful use of Bayesian statistics, the choice of priors is not crucial for Bayesian estimation models with many observations because the influence of the priors diminishes as the number of observations increases.  Thus, the ability of BMM to estimate the percentage of false positives in large samples cannot be explained by the use of priors. It is therefore still not clear how BMM can distinguish between false positives and true positives with low power.

Simulation Studies

The authors report several simulation studies that suggest BMM estimates are close and robust across many scenarios.

“The online supplemental material presents a set of simulation studies that highlight that the model is able to accurately estimate the quantities of interest under a relatively broad range of circumstances”  (p. 1226).

The first set of simulations uses a sample size of N = 500 (n = 250 per condition).  Heterogeneity in effect sizes is simulated with a truncated normal distribution with a standard deviation of .10 (truncated at 2*SD) and effect sizes of d = .45, .30, and .15.  The lowest values are .35, .20, and .05.  With N = 500, these values correspond to  97%, 61%, and 8% power respectively.

d = c(.35,.20,.05); 1-pt(qt(.975,500-2),500-2,d*sqrt(500)/2)

The number of studies was k = 5,000 with half of the studies being false positives (H0) and half being true positives (H1).

Figure 1 shows the Zcurve plot for the simulation with high power (d = .45, power >  97%; median true power = 99.9%).

Sim1.png

The graph shows a bimodal distribution with clear evidence of truncation (the steep drop at z = 1.96 (p = .05, two-tailed) is inconsistent with the distribution of significant z-scores.  The sharp drop from z = 1.96 to 3 shows that there are many studies with non-significant results are missing.  The estimate of unconditional power (called replicability = expected success rate in exact replication studies) is 53%.  This estimate is consistent with the simulation of 50% studies with a probability of success of 5% and 50% of studies with a success probability of 99.9% (.5 * .05 + .5 * .999 = 52.5).

The values below the x-axis show average power for  specific z-scores. A z-score of 2 corresponds roughly to p = .05 and 50% power without selection for significance. Due to selection for significance, the average power is only 9%. Thus the observed power of 50% provides a much inflated estimate of replicability.  A z-score of 3.5 is needed to achieve significance with p < .05, although the nominal p-value for z = 3.5 is p = .0002.  Thus, selection for significance renders nominal p-values meaningless.

The sharp change in power from Z = 3 to Z = 3.5 is due to the extreme bimodal distribution.  While most Z-scores below 3 are from the sampling distribution of H0 (false positives), most Z-scores of 3.5 or higher come from H1 (true positives with high power).

Figure 2 shows the results for the simulation with d = .30.  The results are very similar because d = .30 still gives 92% power.  As a result, replicabilty is nearly as high as in the previous example.

Sim2.png

 

The most interesting scenario is the simulation with low powered true positives. Figure 3 shows the Zcurve for this scenario with an unconditional average power of only 23%.

Sim3.png

It is no longer possible to recognize two sampling distributions and average power increases rather gradually from 18% for z = 2, to 35% for z = 3.5.  Even with this challenging scenario, BMM performed well and correctly estimated the percentage of false positives.   This is surprising because it is easy to generate a similar Zcurve without false positives.

Figure 4 shows a simulation with a mixture distribution but the false positives (d = 0) have been replaced by true positives (d = .06), while the mean for the heterogeneous studies was reduced to from d = .15 to d = .11.  These values were chosen to produce the same average unconditional power (replicability) of 23%.

Sim4.png

I transformed the z-scores into (two-sided) p-values and submitted them to the online BMM app at https://qfgronau.shinyapps.io/bmmsp/ .  I used only k = 1,500 p-values because the server timed me out several times with k = 5,000 p-values.  The estimated percentage of false positives was 24%, with a wide 95% credibility interval ranging from 0% to 48%.   These results suggest that BMM has problems distinguishing between false positives and true positives with low power.   BMM appears to be able to estimate the percentage of false positives correctly when most low z-scores are sampled from H0 (false positives). However, when these z-scores are due to studies with low power, BMM cannot distinguish between false positives and true positives with low power. As a result, the credibility interval is wide and the point estimates are misleading.

BMM.output.png

With k = 1,500 the influence of the priors is negligible.  However, with smaller sample sizes, the priors do have an influence on results and may lead to overestimation and false credibility intervals.  A simulation with k = 200, produced a point estimate of 34% false positives with a very wide CI ranging from 0% to 63%. The authors suggest a sensitivity analysis by changing model parameters. The most crucial parameter is the standard deviation.  Increasing the standard deviation to 2, increases the upper limit of the 95%CI to 75%.  Thus, without good justification for a specific standard deviation, the data provide very little information about the percentage of false positives underlying this Zcurve.

BMM.k200.png

 

For simulations with k = 100, the prior started to bias the results and the CI no longer included the true value of 0% false positives.

BMM.k100

In conclusion, these simulation results show that BMM promises more than it can deliver.  It is very difficult to distinguish p-values sampled from H0 (mean z = 0) and those sampled from H1 with weak evidence (e.g., mean z = 0.1).

In the Challenges and Limitations section, the authors pretty much agree with this assessment of BMM (Gronau et al., 2017, p. 1230).

The procedure does come with three important caveats.

First, estimating the parameters of the mixture model is an inherently difficult statistical problem. ..  and consequently a relatively large number of p values are required for the mixture model to provide informative results. 

A second caveat is that, even when a reasonable number of p values are available, a change in the parameter priors might bring about a noticeably different result.

The final caveat is that our approach uses a simple parametric form to account for the distribution of p values that stem from H1. Such simplicity comes with the risk of model-misspecification.

Practical Implications

Despite the limitations of BMM, the authors applied BMM to several real data.  The most interesting application selected focal hypothesis tests from social priming studies.  Social priming studies have come under attack as a research area with sloppy research methods as well as fraud (Stapel).  Bias tests show clear evidence that published results were obtained with questionable scientific practices (Schimmack, 2017a, 2017b).

The authors analyzed 159 social priming p-values.  The 95%CI for the percentage of false positives ranged from 48% to 88%.  When the standard deviation was increased to 2, the 95%CI increased slightly to 56% to 91%.  However, when the standard deviation was halved, the 95%CI ranged from only 10% to 75%.  These results confirm the authors’ warning that estimates in small sets of studies (k < 200) are highly sensitive to the specification of priors.

What inferences can be drawn from these results about the social priming literature?  A false positive percentage of 10% doesn’t sound so bad.  A false positive percentage of 88% sound terrible. A priori, the percentage is somewhere between 0 and 100%. After looking at the data, uncertainty about the percentage of false positives in the social priming literature remains large.  Proponents will focus on the 10% estimate and critics will use the 88% estimate.  The data simply do not resolve inconsistent prior assumptions about the credibility of discoveries in social priming research.

In short, BMM promises that it can estimate the percentage of false positives in a set of studies, but in practice these estimates are too imprecise and too dependent on prior assumptions to be very useful.

A Zcurve of Social Priming Studies (k = 159)

It is instructive to compare the BMM results to a Zcurve analysis of the same data.

SocialPriming.png

The zcurve graph shows a steep drop and very few z-scores greater than 4, which tend to have a high success rate in actual replication attempts (OSC, 2015).  The average estimated replicability is only 27%.  This is consistent with the more limited analysis of social priming studies in Kahneman’ s Thinking Fast and Slow book (Schimmack, 2017a).

More important than the point estimate is that the 95%CI ranges from 15% to a maximum of 39%.  Thus, even a sample size of 159 studies is sufficient to provide conclusive evidence that these published studies have a low probability of replicating even if it were possible to reproduce the exact conditions again.

These results show that it is not very useful to distinguish between false positives with a replicability of 5% and true positives with a replicability of 6, 10, or 15%.  Good research provides evidence that can be replicated at least with a reasonable degree of statistical power.  Tversky and Kahneman (1971) suggested a minimum of 50% and most social priming studies fail to meet this minimal standard and hardly any studies seem to have been planned with the typical standard of 80% power.

The power estimates below the x-axis show that a nomimal z-score of 4 or higher is required to achieve 50% average power and an actual false positive risk of 5%. Thus, after correcting for deceptive publication practices, most of the seemingly statistically significant results are actually not significant with the common criterion of a 5% risk of a false positive.

The difference between BMM and Zcurve is captured in the distinction between evidence of absence and absence of evidence.  BMM aims to provide evidence of absence (false positives). In contrast, Zcurve has the more modest goal of demonstrating absence (or presence) of evidence.  It is unknown whether any social priming studies could produce robust and replicable effects and under what conditions these effects occur or do not occur.  However, it is not possible to conclude from the poorly designed studies and the selectively reported results that social priming effects are zero.

Conclusion

Zcurve and BMM are both mixture models, but they have different statistical approaches, they have different aims.  They also differ in their ability to provide useful estimates.  Zcurve is designed to estimate average unconditional power to obtain significant results without distinguishing between true positives and false positives.  False positives reduce average power, just like low powered studies, and in reality it can be difficult or impossible to distinguish between a false positive with an effect size of zero and a true positive with an effect size that is negligibly different from zero.

The main problem of BMM is that it treats the nil-hypothesis as an important hypothesis that can be accepted or rejected.  However, this is a logical fallacy.  it is possible to reject an implausible effect sizes (e.g., the nil-hypothesis is probably false if the 95%CI ranges from .8 to  1.2], but it is not possible to accept the nil-hypothesis because there are always values close to 0 that are also consistent with the data.

The problem of BMM is that it contrasts the point-nil-hypothesis with all other values, even if these values are very close to zero.  The same problem plagues the use of Bayes-Factors that compare the point-nil-hypothesis with all other values (Rouder et al., 2009).  A Bayes-Factor in favor of the point nil-hypothesis is often interpreted as if all the other effect sizes are inconsistent with the data.  However, this is a logical fallacy because data that are inconsistent with a specific H1 can be consistent with an alternative H1.  Thus, a BF in favor of H0 can only be interpreted as evidence against a specific H1, but never as evidence that the nil-hypothesis is true.

To conclude, I have argued that it is more important to estimate the replicability of published results than to estimate the percentage of false positives.  A literature with 100% true positives and average power of 10% is no more desirable than a literature with 50% false positives and 50% true positives with 20% power.  Ideally, researchers should conduct studies with 80% power and honest reporting of statistics and failed replications should control the false discovery rate.  The Zcurve for social priming studies shows that priming researchers did not follow these basic and old principles of good science.  As a result, decades of research are worthless and Kahneman was right to compare social priming research to a train wreck because the conductors ignored all warning signs.

 

 

 

Charles Stangor’s Failed Attempt to Predict the Future

Charles Stangor, Replicability, Replication Crisis, Stangor, Zcurve

Background

It is 2018, and 2012 is a faint memory.  So much has happened in the word and in
psychology over the past six years.

Two events rocked Experimental Social Psychology (ESP) in the year 2011 and everybody was talking about the implications of these events for the future of ESP.

First, Daryl Bem had published an incredible article that seemed to suggest humans, or at least extraverts, have the ability to anticipate random future events (e.g., where an erotic picture would be displayed).

Second, it was discovered that Diederik Stapel had fabricated data for several articles. Several years later, over 50 articles have been retracted.

Opinions were divided about the significance of these two events for experimental social psychology.  Some psychologists suggested that these events are symptomatic of a bigger crisis in social psychology.  Others considered these events as exceptions with little consequences for the future of experimental social psychology.

In February 2012, Charles Stangor tried to predict how these events will shape the future of experimental social psychology in an essay titled “Rethinking my Science”

How will social and personality psychologists look back on 2011? With pride at having continued the hard work of unraveling the mysteries of human behavior, or with concern that the only thing that is unraveling is their discipline?

Stangor’s answer is clear.

“Although these two events are significant and certainly deserve our attention, they are flukes rather than game-changers.”

He describes Bem’s article as a “freak event” and Stapel’s behavior as a “fluke.”

“Some of us probably do fabricate data, but I imagine the numbers are relatively few.”

Stangor is confident that experimental social psychology is not really affected by these two events.

As shocking as they are, neither of these events create real problems for social psychologists

In a radical turn, Stangor then suggests that experimental social psychology will change, but not in response to these events, but in response to three other articles.

But three other papers published over the past two years must completely change how we think about our field and how we must conduct our research within it. And each is particularly important for me, personally, because each has challenged a fundamental assumption that was part of my training as a social psychologist.

Student Samples

The first article is a criticism of experimental social psychology for relying too much on first-year college students as participants (Heinrich, Heine, & Norenzayan, 2010).  Looking back, there is no evidence that US American psychologists have become more global in their research interests. One reason is that social phenomena are sensitive to the cultural context and for Americans it is more interesting to study how online dating is changing relationships than to study arranged marriages in more traditional cultures. There is nothing wrong with a focus on a particular culture.  It is not even clear that research article on prejudice against African Americans were supposed to generalize to the world (how would this research apply to African countries where the vast majority of citizens are black?).

The only change that occurred was not in response to Heinrich et al.’s (2010) article, but in response to technological changes that made it easier to conduct research and pay participants online.  Many social psychologists now use the online service Mturk to recruit participants.

Thus, I don’t think this article significantly changed experimental social psychology.

Decline Effect 

The second article with the title (“The Truth Wears Off“) was published in the weekly magazine the New Yorker.  It made the ridiculous claim that true effects become weaker or may even disappear over time.

The basic phenomenon is that observed findings in the social and biological sciences weaken with time. Effects that are easily replicable at first become less so every day. Drugs stop working over time the same way that social psychological phenomena become more and more elusive. The “the decline effect” or “the truth wears off effect,” is not easy to dismiss, although perhaps the strength of the decline effect will itself decline over time.

The assumption that the decline effect applies to real effects is no more credible than Bem’s claims of time-reversed causality.   I am still waiting for the effect of eating cheesecake on my weight (a biological effect) to wear off. My bathroom scale tells me it is not.

Why would Stangor believe in such a ridiculous idea?  The answer is that he observed it many times in his own work.

Frankly I have difficulty getting my head around this idea (I’m guessing others do too) but it is nevertheless exceedingly troubling. I know that I need to replicate my effects, but am often unable to do it. And perhaps this is part of the reason. Given the difficulty of replication, will we continue to even bother? And what becomes of our research if we do even less replicating than we do now? This is indeed a problem that does not seem likely to go away soon. 

In hindsight, it is puzzling that Stangor misses the connection between Bem’s (2011) article and the decline effect.   Bem published 9 successful results with p < .05.  This is not a fluke. The probability to get lucky 9 times in a row with a probability of just 5% for a single event is very very small (less than 1 in a billion attempts).  It is not a fluke. Bem also did not fabricate data like Stapel, but he falsified data to present results that are too good to be true (Definitions of Research Misconduct).  Not surprisingly, neither he nor others can replicate these results in transparent studies that prevent the use of QRPs (just like paranormal phenomena like spoon bending can not be replicated in transparent experiments that prevent fraud).

The decline effect is real, but it is wrong to misattribute it to a decline in the strength of a true phenomenon.  The decline effect occurs when researchers use questionable research practices (John et al., 2012) to fabricate statistically significant results.  Questionable research practices inflate “observed effect sizes” [a misnomer because effects cannot be observed]; that is, the observed mean differences between groups in an experiment.  Unfortunately, social psychologists do not distinguish between “observed effects sizes” and true or population effect sizes. As a result, they believe in a mysterious force that can reduce true effect sizes when sampling error moves mean differences in small samples around.

In conclusion, the truth does not wear off because there was no truth to begin with. Bem’s (2011) results did not show a real effect that wore off in replication studies. The effect was never there to begin with.

P-Hacking

The third article mentioned by Stangor did change experimental social psychology.  In this article, Simmons, Nelson, and Simonsohn (2011) demonstrate the statistical tricks experimental social psychologists have used to produce statistically significant results.  They call these tricks, p-hacking.  All methods of p-hacking have one common feature. Researchers conduct mulitple statistical analysis and check the results. When they find a statistically significant result, they stop analyzing the data and report the significant result.  There is nothing wrong with this practice so far, but it essentially constitutes research misconduct when the result is reported without fully disclosing how many attempts were made to get it.  The failure to disclose all attempts is deceptive because the reported result (p < .05) is only valid if a researcher collected data and then conducted a single test of a hypothesis (it does not matter whether this hypothesis was made before or after data collection).  The point is that at the moment a researcher presses a mouse button or a key on a keyboard to see a p-value,  a statistical test occurred.  If this p-value is not significant and another test is run to look at another p-value, two tests are conducted and the risk of a type-I error is greater than 5%. It is no longer valid to claim p < .05, if more than one test was conducted.  With extreme abuse of the statistical method (p-hacking), it is possible to get a significant result even with randomly generated data.

In 2010, the Publication Manual of the American Psychological Association advised researchers that “omitting troublesome observations from reports to present a more convincing story is also prohibited” (APA).  It is telling that Stangor does not mention this section as a game-changer, because it has been widely ignored by experimental psychologists until this day.  Even Bem’s (2011) article that was published in an APA journal violated this rule, but it has not been retracted or corrected so far.

The p-hacking article had a strong effect on many social psychologists, including Stangor.

Its fundamental assertions are deep and long-lasting, and they have substantially affected me. 

Apparently, social psychologists were not aware that some of their research practices undermined the credibility of their published results.

Although there are many ways that I take the comments to heart, perhaps most important to me is the realization that some of the basic techniques that I have long used to collect and analyze data – techniques that were taught to me by my mentors and which I have shared with my students – are simply wrong.

I don’t know about you, but I’ve frequently “looked early” at my data, and I think my students do too. And I certainly bury studies that don’t work, let alone fail to report dependent variables that have been uncooperative. And I have always argued that the researcher has the obligation to write the best story possible, even if may mean substantially “rewriting the research hypothesis.” Over the years my students have asked me about these practices (“What do you recommend, Herr Professor?”) and I have
routinely, but potentially wrongly, reassured them that in the end, truth will win out. 

Although it is widely recognized that many social psychologists p-hacked and buried studies that did not work out,  Stangor’s essay remains one of the few open admissions that these practices were used, which were not considered unethical, at least until 2010. In fact, social psychologists were trained that telling a good story was essential for social psychologists (Bem, 2001).

In short, this important paper will – must – completely change the field. It has shined a light on the elephant in the room, which is that we are publishing too many Type-1 errors, and we all know it.

Whew! What a year 2011 was – let’s hope that we come back with some good answers to these troubling issues in 2012.

In hindsight Stangor was right about the p-hacking article. It has been cited over 1,000 times so far and the term p-hacking is widely used for methods that essentially constitute a violation of research ethics.  P-values are only meaningful if all analyses are reported and failures to disclose analyses that produced inconvenient non-significant results to tell a more convincing story constitutes research misconduct according to the guidelines of APA and the HHS.

Charles Stangor’s Z-Curve

Stangor’s essay is valuable in many ways.  One important contribution is the open admission to the use of QRPs before the p-hacking article made Stangor realize that doing so was wrong.   I have been working on statistical methods to reveal the use of QRPs.  It is therefore interesting to see the results of this method when it is applied to data by a researcher who used QRPs.

stangor.png

This figure (see detailed explanation here) shows the strength of evidence (based on test statistics like t and F-values converted into z-scores in Stangor’s articles. The histogram shows a mode at 2, which is just significant (z = 1.96 ~ p = .05, two-tailed).  The steep drop on the left shows that Stangor rarely reported marginally significant results (p = .05 to .10).  It also shows the use of questionable research practices because sampling error should produce a larger number of non-significant results than are actually observed. The grey line provides a vague estimate of the expected proportion of non-significant results. The so called file-drawer (non-significant results that are not reported) is very large.  It is unlikely that so many studies were attempted and not reported. As Stangor mentions, he also used p-hacking to get significant results.  P-hacking can produce just significant results without conducting many studies.

In short, the graph is consistent with Stangor’s account that he used QRPs in his research, which was common practice and even encouraged, and did not violate any research ethics code of the times (Bem, 2001).

The graph also shows that the significant studies have an estimated average power of 71%.  This means any randomly drawn statistically significant result from Stangor’s articles has a 71% chance of producing a significant result again, if the study and the statistical test were replicated exactly (see Brunner & Schimmack, 2018, for details about the method).  This average is not much below the 80% value that is considered good power.

There are two caveats with the 71% estimate. One caveat is that this graph uses all statistical tests that are reported, but not all of these tests are interesting. Other datasets suggest that the average for focal hypothesis tests is about 20-30 percentage points lower than the estimate for all tests. Nevertheless, an average of 71% is above average for social psychology.

The second caveat is that there is heterogeneity in power across studies. Studies with high power are more likely to produce really small p-values and larger z-scores. This is reflected in the estimates below the x-axis for different segments of studies.  The average for studies with just significant results (z = 2 to 2.5) is only 49%.  It is possible to use the information from this graph to reexamine Stangor’s articles and to adjust nominal p-values.  According to this graph p-values in the range between .05 and .01 would not be significant because 50% power corresponds to a p-value of .05. Thus, all of the studies with a z-score of 2.5 or less (~ p > .01) would not be significant after correcting for the use of questionable research practices.

The main conclusion that can be drawn from this analysis is that the statistical analysis of Stangor’s reported results shows convergent validity with the description of his research practices.  If test statistics by other researchers show a similar (or worse) distribution, it is likely that they also used questionable research practices.

Charles Stangor’s Response to the Replication Crisis 

Stangor was no longer an active researcher when the replication crisis started. Thus, it is impossible to see changes in actual research practices.  However, Stangor co-edited a special issue for the Journal of Experimental Social Psychology on the replication crisis.

The Introduction mentions the p-hacking article.

At the same time, the empirical approaches adopted by social psychologists leave room for practices that distort or obscure the truth (Hales, 2016-in this issue; John, Loewenstein, & Prelec, 2012; Simmons, Nelson, & Simonsohn, 2011)

and that

social psychologists need to do some serious housekeeping in order to progress
as a scientific enterprise.

It quotes, Dovidio to claim that social psychologists are

lucky to have the problem. Because social psychologists are rapidly developing new approaches and techniques, our publications will unavoidably contain conclusions that are uncertain, because the potential limitations of these procedures are not yet known. The trick then is to try to balance “new” with “careful.

It also mentions the problem of fabricating stories by hiding unruly non-significant results.

The availability of cheap data has a downside, however,which is that there is little cost in omitting data that contradict our hypotheses from our manuscripts (John et al., 2012). We may bury unruly data because it is so cheap and plentiful. Social psychologists justify this behavior, in part, because we think conceptually. When a manipulation fails, researchers may simply argue that the conceptual variable was not created by that particular manipulation and continue to seek out others that will work. But when a study is eventually successful,we don’t know if it is really better than the others or if it is instead a Type I error. Manipulation checks may help in this regard, but they are not definitive (Sigall &Mills, 1998).

It also mentioned file-drawers with unsuccessful studies like the one shown in the Figure above.

Unpublished studies likely outnumber published studies by an order of magnitude. This is wasteful use of research participants and demoralizing for social psychologists and their students.

It also mentions that governing bodies have failed to crack down on the use of p-hacking and other questionable practices and the APA guidelines are not mentioned.

There is currently little or no cost to publishing questionable findings

It foreshadows calls for a more stringent criterion of statistical significance, known as the p-value wars (alpha  = .05 vs. alpha = .005 vs. justify your alpha vs. abandon alpha)

Researchers base statistical analyses on the standard normal distribution but the actual tails are probably bigger than this approach predicts. It is clear that p b .05 is not enough to establish the credibility of an effect. For example, in the Reproducibility Project (Open Science Collaboration, 2015), only 18% of studies with a p-value greater than .04 replicated whereas 63% of those with a p-value less than .001 replicated. Perhaps we should require, at minimum, p <  .01 

It is not clear, why we should settle for p < .01, if only 63% of results replicated with p < .001. Moreover, it ignores that a more stringent criterion for significance also increases the risk of type-II error (Cohen).  It also ignores that only two studies are required to reduce the risk of a type-I error from .05 to .05*.05 = .0025.  As many articles in experimental social psychology are based on multiple cheap studies, the nominal type-I error rate is well below .001.  The real problem is that the reported results are not credible because QRPs are used (Schimmack, 2012).  A simple and effective way to improve experimental social psychology would be to enforce the APA ethics guidelines and hold violators of these rules accountable for their actions.  However, although no new rules would need to be created, experimental social psychologists are unable to police themselves and continue to use QRPs.

The Introduction ignores this valid criticism of multiple study and continues to give the misleading impression that more studies translate into more replicable results.  However, the Open-Science Collaboration reproducibility project showed no evidence that long, multiple-study articles reported more replicable results than shorter articles in Psychological Science.

In addition, replication concerns have mounted with the editorial practice of publishing short papers involving a single, underpowered study demonstrating counterintuitive results (e.g., Journal of Experimental Social Psychology; Psychological Science; Social Psychological and Personality Science). Publishing newsworthy results quickly has benefits,
but also potential costs (Ledgerwood & Sherman, 2012), including increasing Type 1 error rates (Stroebe, 2016-in this issue). 

Once more, the problem is dishonest reporting of results.  A risky study can be published and a true type-I error rate of 20% informs readers that there is a high risk of a false positive result. In contrast, 9 studies with a misleading type-I error rate of 5% violate the implicit assumptions that readers can trust a scientific research article to report the results of an objective test of a scientific question.

But things get worse.

We do, of course, understand the value of replication, and publications in the premier social-personality psychology journals often feature multiple replications of the primary findings. This is appropriate, because as the number of successful replications increases, our confidence in the finding also increases dramatically. However, given the possibility
of p-hacking (Head, Holman, Lanfear, Kahn, & Jennions, 2015; Simmons et al., 2011) and the selective reporting of data, replication is a helpful but imperfect gauge of whether an effect is real. 

Just like Stangor dismissed Bem’s mulitple-study article in JPSP as a fluke that does not require further attention, he dismisses evidence that QRPs were used to p-hack other multiple study articles (Schimmack, 2012).  Ignoring this evidence is just another violation of research ethics. The data that are being omitted here are articles that contradict the story that an author wants to present.

And it gets worse.

Conceptual replications have been the field’s bread and butter, and some authors of the special issue argue for the superiority of conceptual over exact replications (e.g. Crandall & Sherman, 2016-in this issue; Fabrigar and Wegener, 2016–in this issue; Stroebe, 2016-in this issue).  The benefits of conceptual replications are many within social psychology, particularly because they assess the robustness of effects across variation in methods, populations, and contexts. Constructive replications are particularly convincing because they directly replicate an effect from a prior study as exactly as possible in some conditions but also add other new conditions to test for generality or limiting conditions (Hüffmeier, 2016-in this issue).

Conceptual replication is a euphemism for story telling or as Sternberg calls it creative HARKing (Sternberg, in press).  Stangor explained earlier how an article with several conceptual replication studies is constructed.

I certainly bury studies that don’t work, let alone fail to report dependent variables that have been uncooperative. And I have always argued that the researcher has the obligation to write the best story possible, even if may mean substantially “rewriting the research hypothesis.”

This is how Bem advised generations of social psychologists to write articles and that is how he wrote his 2011 article that triggered awareness of the replicability crisis in social psychology.

There is nothing wrong with doing multiple studies and to examine conditions that make an effect stronger or weaker.  However, it is psuedo-science if such a program of research reports only successful results because reporting only successes renders statistical significance meaningless (Sterling, 1959).

The miraculous conceptual replications of Bem (2011) are even more puzzling in the context of social psychologists conviction that their effects can decrease over time (Stangor, 2012) or change dramatically from one situation to the next.

Small changes in social context make big differences in experimental settings, and the same experimental manipulations create different psychological states in different times, places, and research labs (Fabrigar andWegener, 2016–in this issue). Reviewers and editors would do well to keep this in mind when evaluating replications. 

How can effects be sensitive to context and the success rate in published articles is 95%?

And it gets worse.

Furthermore, we should remain cognizant of the fact that variability in scientists’ skills can produce variability in findings, particularly for studies with more complex protocols that require careful experimental control (Baumeister, 2016-in this issue). 

Baumeister is one of the few other social psychologists who has openly admitted not disclosing failed studies.  He also pointed out that in 2008 this practice did not violate APA standards.  However, in 2016 a major replication project failed to replicate the ego-depletion effect that he first “demonstrated” in 1998.  In response to this failure, Baumeister claimed that he had produced the effect many times, suggesting that he has some capabilities that researchers who fail to show the effect lack (in his contribution to the special issue in JESP he calls this ability “flair”).  However, he failed to mention that many of his attempts failed to show the effect and that his high success rate in dozens of articles can only be explained by the use of QRPs.

While there is ample evidence for the use of QRPs, there is no empirical evidence for the claim that research expertise matters.  Moreover, most of the research is carried out by undergraduate students supervised by graduate students and the expertise of professors is limited to designing studies and not to actually carrying out studies.

In the end, the Introduction also comments on the process of correcting mistakes in published articles.

Correctors serve an invaluable purpose, but they should avoid taking an adversarial tone. As Fiske (2016–this issue) insightfully notes, corrective articles should also
include their own relevant empirical results — themselves subject to
correction.

This makes no sense. If somebody writes an article and claims to find an interaction effect based on a significant result in one condition and a non-significant result in another condition, the article makes a statistical mistake (Gelman & Stern, 2005). If a pre-registration contains the statement that an interaction is predicted and a published article claims an interaction is not necessary, the article misrepresents the nature of the preregistration.  Correcting mistakes like this is necessary for science to be a science.  No additional data are needed to correct factual mistakes in original articles (see, e.g., Carlsson, Schimmack, Williams, & Bürkner, 2017).

Moreover, Fiske has been inconsistent in her assessment of psychologists who have been motivated by the events of 2011 to improve psychological science.  On the one hand, she has called these individuals “method terrorists” (2016 review).  On the other hand, she suggests that psychologists should welcome humiliation that may result from the public correction of a mistake in a published article.

Conclusion

In 2012, Stangor asked “How will social and personality psychologists look back on 2011?” Six years later, it is possible to provide at least a temporary answer. There is no unified response.

The main response by older experimental social psychologist has been denial along Stangor’s initial response to Stapel and Bem.  Despite massive replication failures and criticism, including criticism by Noble Laureate Daniel Kahneman, no eminent social psychologists has responded to the replication crisis with an admission of mistakes.  In contrast, the list of eminent social psychologists who stand by their original findings despite evidence for the use of QRPs and replication failures is long and is growing every day as replication failures accumulate.

The response by some younger social psychologists has been to nudge social psychologists slowly towards improving their research methods, mainly by handing out badges for preregistrations of new studies.  Although preregistration makes it more difficult to use questionable research practices, it is too early to see how effective preregistration is in making published results more credible.  Another initiative is to conduct replication studies. The problem with this approach is that the outcome of replication studies can be challenged and so far these studies have not resulted in a consensual correction in the scientific literature. Even articles that reported studies that failed to replicate continue to be cited at a high rate.

Finally, some extremists are asking for more radical changes in the way social psychologists conduct research, but these extremists are dismissed by most social psychologists.

It will be interesting to see how social psychologists, funding agencies, and the general public will look back on 2011 in 2021.  In the meantime, social psychologists have to ask themselves how they want to be remembered and new investigators have to examine carefully where they want to allocate their resources.  The published literature in social psychology is a mine field and nobody knows which studies can be trusted or not.

I don’t know about you, but I am looking forward to reading the special issues in 2021 in celebration of the 10-year anniversary of Bem’s groundbreaking or should I saw earth-shattering publication of “Feeling the Future.”

Visual Inspection of Strength of Evidence: P-Curve vs. Z-Curve

P-Curve, Pcurve, Power, Powergraph, Strength of Evidence, Z-Curve, Zcurve

Statistics courses often introduce students to a bewildering range of statistical test.  They rarely point out how test statistics are related.  For example, although t-tests may be easier to understand than F-tests, every t-test could be performed as an F-test and the F-value in the F-test is simply the square of the t-value (t^2 or t*t).

At an even more conceptual level, all test statistics are ratios of the effect size (ES) and the amount of sampling error (ES).   The ratio is sometimes called the signal (ES) to noise (ES) ratio.  The higher the signal to noise ratio (ES/SE), the stronger the observed results deviate from the hypothesis that the effect size is zero.  This hypothesis is often called the null-hypothesis, but this terminology has created some confusing.  It is also sometimes called the nil-hypothesis the zero-effect hypothesis or the no-effect hypothesis.  Most important, the test-statistic is expected to average zero if the same experiment could be replicated a gazillion times.

The test statistics of statistical tests cannot be directly compared.  A t-value of 2 in a study with N = 10 participants provides weaker evidence against the null-hypothesis than a z-score of 1.96.  and an F-value of 4 with df(1,40) provides weaker evidence than an F(10,200) = 4 result.  It is only possible to compare test values directly that have the same sampling distribution (z with z, F(1,40) with F(1,40), etc.).

There are three solutions to this problem. One solution is to use effect sizes as the unit of analysis. This is useful if the aim is effect size estimation.  Effect size estimation has become the dominant approach in meta-analysis.  This blog post is not about effect size estimation.  I just mention it because many readers may be familiar with effect size meta-analysis, but not familiar with meta-analysis of test statistics that reflect the ratio of effect size and sampling error (Effect size meta-analysis: unit = ES; Test Statistic Meta-Analysis: unit ES/SE).

P-Curve

There are two approaches to standardize test statistics so that they have a common unit of measurement.  The first approach goes back to Ronald Fisher, who is considered the founder of modern statistics for researchers.  Following Fisher it is common practice to convert test-statistics into p-values (for this blog post assumes that you are familiar with p-values).   P-values have the same meaning independent of the test statistic that was used to compute them.   That is, p = .05 based on a z-test, t-test, or an F-test provide equally strong evidence against the null-hypothesis (Bayesians disagree, but that is a different story).   The use of p-values as a common metric to examine strength of evidence (evidential value) was largely forgotten, until Simonsohn, Simmons, and Nelson (SSN) used p-values to develop a statistical tool that takes publication bias and questionable research practices into account.  This statistical approach is called p-curve.  P-curve is a family of statistical methods.  This post is about the p-curve plot.

A p-curve plot is essentially a histogram of p-values with two characteristics. First, it only shows significant p-values (p < .05, two-tailed).  Second, it plots the p-values between 0 and .05 with 5 bars.  The Figure shows a p-curve for Motyl et al.’s (2017) focal hypothesis tests in social psychology.  I only selected t-test and F-tests from studies with between-subject manipulations.

p.curve.motyl

The main purpose of a p-curve plot is to examine whether the distribution of p-values is uniform (all bars have the same height).  It is evident that the distribution for Motyl et al.’s data is not uniform.  Most of the p-values fall into the lowest range between 0 and .01. This pattern is called “rigth-skewed.”  A right-skewed plot shows that the set of studies has evidential value. That is, some test statistics are based on non-zero effect sizes.  The taller the bar on the left is, the greater the proportion of studies with an effect.  Importantly, meta-analyses of p-values do not provide information about effect sizes because p-values take effect size and sampling error into account.

The main inference that can be drawn from a visual inspection of a p-curve plot is how unlikely it is that all significant results are false positives; that is, the p-value is below .05 (statistically significant), but this strong deviation from 0 was entirely due to sampling error, while the true effect size is 0.

The next Figure also shows a plot of p-values.  The difference is that it shows the full range of p-values and that it differentiates more between p-values because p = .09 provides weaker evidence than p = .0009.

all.p.curve.motyl.png

The histogram shows that most p-values are below p < .001.  It also shows very few non-significant results.  However, this plot is not more informative than the actual p-curve plot. The only conclusion that is readily visible is that the distribution is not uniform.

The main problem with p-value plots is that p-values do not have interval scale properties.  This means, the difference between p = .4 and p = .3 is not the same as the difference between p = .10 and p = .00 (e.g., .001).

Z-Curve  

Stouffer developed an alternative method to Fisher’s p-value meta-analysis.  Every p-value can be transformed into a z-scores that corresponds to a particular p-value.  It is important to distinguish between one-sided and two-sided p-values.  The transformation requires the use of one-sided p-values, which can be obtained by simply dividing a two-sided p-value by 2.  A z-score of -1.96 corresponds to a one-sided p-value of 0.025 and a z-score of 1.96 corresponds to a one-sided p-values of 0.025.  In a two sided test, the sign no longer matters and the two p-values are added to yield 0.025 + 0.025 = 0.05.

In a standard meta-analysis, we would want to use one-sided p-values to maintain information about the sign.  However, if the set of studies examines different hypothesis (as in Motyl et al.’s analysis of social psychology in general) the sign is no longer important.   So, the transformed two-sided p-values produce absolute (only positive) z-scores.

The formula in R is Z = -qnorm(p/2)   [p = two.sided p-value]

For very strong evidence this formula creates problems. that can be solved by using the log.P=TRUE option in R.

Z = -qnorm(log(p/2), log.p=TRUE)

p.to.z.transformation.png

The plot shows the relationship between z-scores and p-values.  While z-scores are relatively insensitive to variation in p-values from .05 to 1, p-values are relatively insensitive to variation in z-scores from 2 to 15.

only.sig.p.to.z.transformation

The next figure shows the relationship only for significant p-values.  Limiting the distribution of p-values does not change the fact that p-values and z-values have very different distributions and a non-linear relationship.

The advantage of using (absolute) z-scores is that z-scores have ratio scale properties.  A z-score of zero has real meaning and corresponds to the absence of evidence for an effect; the observed effect size is 0.  A z-score of 2 is twice as strong as a z-score of 1. For example, given the same sampling error the effect size for a z-score of 2 is twice as large as the effect size for a z-score of 1 (e.g., d = .2, se = .2, z = d/se = 1,  d = 4, se = .2, d/se = 2).

It is possible to create the typical p-curve plot with z-scores by selecting only z-scores above z = 1.96. However, this graph is not informative because the null-hypothesis does not predict a uniform distribution of z-scores.   For z-values the central tendency of z-values is more important.  When the null-hypothesis is true, p-values have a uniform distribution and we would expect an equal number of p-values between 0 and 0.025 and between 0.025 and 0.050.   A two-sided p-value of .025 corresponds to a one-sided p-value of 0.0125 and the corresponding z-value is 2.24

p = .025
-qnorm(log(p/2),log.p=TRUE)
[1] 2.241403

Thus, the analog to a p-value plot is to examine how many significant z-scores fall into the region from 1.96 to 2.24 versus the region with z-values greater than 2.24.

z.curve.plot1.png

The histogram of z-values is called z-curve.  The plot shows that most z-values are in the range between 1 and 6, but the histogram stretches out to 20 because a few studies had very high z-values.  The red line shows z = 1.96. All values on the left are not significant with alpha = .05 and all values on the right are significant (p < .05).  The dotted blue line corresponds to p = .025 (two tailed).  Clearly there are more z-scores above 2.24 than between 1.96 and 2.24.  Thus, a z-curve plot provides the same information as a p-curve plot.  The distribution of z-scores suggests that some significant results reflect true effects.

However, a z-curve plot provides a lot of additional information.  The next plot removes the long tail of rare results with extreme evidence and limits the plot to z-scores in the range between 0 and 6.  A z-score of six implies a signal to noise ratio of 6:1 and corresponds to a p-value of p = 0.000000002 or 1 out of 2,027,189,384 (~ 2 billion) events. Even particle physics settle for z = 5 to decide that an effect was observed if it is so unlikely for a test result to occur by chance.

> pnorm(-6)*2
[1] 1.973175e-09

Another addition to the plot is to include a line that identifies z-scores between 1.65 and 1.96.  These z-scores correspond to two-sided p-values between .05 and .10. These values are often published as weak but sufficient evidence to support the inference that a (predicted) effect was detected. These z-scores also correspond to p-values below .05 in one-sided tests.

z.curve.plot2

A major advantage of z-scores over p-values is that p-values are conditional probabilities based on the assumption that the null-hypothesis is true, but this hypothesis can be safely rejected with these data.  So, the actual p-values are not important because they are conditional on a hypothesis that we know to be false.   It is like saying, I would be a giant if everybody else were 1 foot tall (like Gulliver in Lilliput), but everybody else is not 1 foot tall and I am not a giant.

Z-scores are not conditioned on any hypothesis. They simply show the ratio of the observed effect size and sampling error.  Moreover, the distribution of z-scores tell us something about the ratio of the true effect sizes and sampling error.  The reason is that sampling error is random and like any random variable has a mean of zero.  Therefore, the mode, median, or mean of a z-curve plot tells us something about ratio of the true effect sizes and sampling error.  The more the center of a distribution is shifted to the right, the stronger is the evidence against the null-hypothesis.  In a p-curve plot, this is reflected in the height of the bar with p-values below .01 (z > 2.58), but a z-curve plot shows the actual distribution of the strength of evidence and makes it possible to see where the center of a distribution is (without more rigorous statistical analyses of the data).

For example, in the plot above it is not difficult to see the mode (peak) of the distribution.  The most common z-values are between 2 and 2.2, which corresponds to p-values of .046 (pnorm(-2.2)*2) and .028 (pnorm(-2.2)*2).   This suggests that the modal study has a ratio of 2:1 for effect size over sampling error.

The distribution of z-values does not look like a normal distribution. One explanation for this is that studies vary in sampling errors and population effect sizes.  Another explanation is that the set of studies is not a representative sample of all studies that were conducted.   It is possible to test this prediction by trying to fit a simple model to the data that assumes representative sampling of studies (no selection bias or p-hacking) and that assumes that all studies have the same ratio of population effect size over sampling error.   The median z-score provides an estimate of the center of the sampling distribution.  The median for these data is z = 2.56.   The next picture shows the predicted sampling distribution of this model, which is an approximately normal distribution with a folded tail.

 

z.curve.plot3

A comparison of the observed and predicted distribution of z-values shows some discrepancies. Most important is that there are too few non-significant results.  This observation provides evidence that the results are not a representative sample of studies.  Either non-significant results were not reported or questionable research practices were used to produce significant results by increasing the type-I error rate without reporting this (e.g., multiple testing of several DVs, or repeated checking for significance during the course of a study).

It is important to see the difference between the philosophies of p-curve and z-curve. p-curve assumes that non-significant results provide no credible evidence and discards these results if they are reported.  Z-curve first checks whether non-significant results are missing.  In this way, p-curve is not a suitable tool for assessing publication bias or other problems, whereas even a simple visual inspection of z-curve plots provides information about publication bias and questionable research practices.

z.curve.plot4.png

The next graph shows a model that selects for significance.  It no longer attempts to match the distribution of non-significant results.  The objective is only to match the distribution of significant z-values.  You can do this by hand and simply try out different values for the center of the normal distribution.  The lower the center, the more z-scores are missing because they are not significant.  As a result, the density of the predicted curve needs to be adjusted to reflect the fact that some of the area is missing.

center.z = 1.8  #pick a value
z = seq(0,6,.001)  #create the range of z-values
y = dnorm(z,center.z,1) + dnorm(z,-center.z,1)  # get the density for a folded normal
y2 = y #duplicate densities
y2[x < 1.96] = 0   # simulate selection bias, density for non-significant results is zero
scale = sum(y2)/sum(y)  # get the scaling factor so that area under the curve of only significant results is 1.
y = y / scale   # adjust the densities accordingly

# draw a histogram of z-values
# input is  z.val.input
# example; z.val.input = rnorm(1000,2)
hist(z.val.input,freq=FALSE,xlim=c(0,6),ylim=c(0,1),breaks=seq(0,20,.2), xlab=””,ylab=”Density”,main=”Z-Curve”)

abline(v=1.96,col=”red”)   # draw the line for alpha = .05 (two-tailed)
abline(v=1.65,col=”red”,lty=2)  # draw marginal significance (alpha = .10 (two-tailed)

par(new=TRUE) #command to superimpose next plot on histogram

# draw the predicted sampling distribution
plot(x,y,type=”l”,lwd=4,ylim=c(0,1),xlim=c(0,6),xlab=”(absolute) z-values”,ylab=””)

Although this model fits the data better than the previous model without selection bias, it still has problems fitting the data.  The reason is that there is substantial heterogeneity in the true strength of evidence.  In other words, the variability in z-scores is not just sampling error but also variability in sampling errors (some studies have larger samples than others) and population effect sizes (some studies examine weak effects and others examine strong effects).

Jerry Brunner and I developed a mixture model to fit a predicted model to the observed distribution of z-values.  In a nutshell the mixture model has multiple (folded) normal distributions.  Jerry’s z-curve lets the center of the normal distribution move around and give different weights to them.  Uli’s z-curve uses fixed centers one standard deviation apart (0,1,2,3,4,5 & 6) and uses different weights to fit the model to the data.  Simulation studies show that both methods work well.  Jerry’s method works a bit better if there is little variability and Uli’s method works a bit better with large variability.

The next figure shows the result for Uli’s method because the data have large variability.

z.curve.plot5

The dark blue line in the figure shows the density distribution for the observed data.  A density distribution assigns densities to an observed distribution that does not fit a mathematical sampling distribution like the standard normal distribution.   We use the Kernel Density Estimation method implemented in the R base package.

The grey line shows the predicted density distribution based on Uli’s z-curve method.  The z-curve plot makes it easy to see the fit of the model to the data, which is typically very good.  The model result of the model is the weighted average of the true power that corresponds to the center of the simulated normal distributions.  For this distribution,  the weighted average is 48%.

The 48% estimate can be interpreted in two ways.  First, it means that if researchers randomly sampled from the set of studies in social psychology and were able to exactly reproduce the original study (including sample size),  they have a probability of 48% to replicate a significant result with alpha = .05.  The complementary interpretation is that if researchers were successful in replicating all studies exactly,  the reproducibility project is expected to produce 48% significant results and 52% non-significant results.  Because average power of studies predicts the success of exact replication studies, Jerry and I refer to the average power of studies that were selected for significance replicability.  Simulation studies show that our z-curve methods have good large sample accuracy (+/- 2%) and we adjust for the small estimation bias in large samples by computing a conservative confidence interval that adds 2% to the upper limit and 2% to the lower limit.

Below is the R-Code to obtain estimates of replicability from a set of z-values using Uli’s method.

<<<Download Zcurve R.Code>>>

Install R.Code on your computer, then run from anywhere with the following code

location =  <user folder>  #provide location information where z-curve code is stored
source(paste0(location,”fun.uli.zcurve.sharing.18.1.R”))  #read the code
run.zcurve(z.val.input)  #get z-curve estimates with z-values as input