Aleksi Reito writes:

The statement below was included in a recent issue of Annals of Surgery:

But, as 80% power is difficult to achieve in surgical studies, we argue that the CONSORT and STROBE guidelines should be modified to include the disclosure of power—even if less than 80%—with the given sample size and effect size observed in that study.

It is the highest ranking journal in the field of surgery. I find it worrying that they suggest calculating post-hoc power.

I agree. This is a well known error; see references here, where we write:

The idea that published effect-size estimates tend to be too large, essentially because of publication bias, is not new (Hedges, 1984; Lane & Dunlap, 1978; for a more recent example, also see Button et al., 2013). . . .

After data have been collected, and a result is in hand, statistical authorities commonly recommend against performing power calculations (see, e.g., Goodman & Berlin, 1994; Lenth, 2007; Senn, 2002).

It’s fine to estimate power (or, more generally, statistical properties of estimates) after the data have come in—but only only only only only if you do this based on a scientifically grounded assumed effect size. One should not not not not not estimate the power (or other statistical properties) of a study based on the “effect size observed in that study.” That’s just terrible, and it’s too bad that the Annals of Surgery is ignoring a literature that goes back at least to 1994 (and I’m sure earlier) that warns against this.

Reito continues:

I still can´t understand how it is possible that authors suggest a revision to CONSORT and STROBE guidelines by including an assessment of post-hoc power and this gets published in the highest ranking surgical journal. They try to tackle the issues with reproducibility but show a complete lack of understanding in the basic statistical concepts. I look forward the discussion on this matter.

I too look forward to this discussion. Hey, Annals of Surgery, whassup?

P.S. I guess I could write a letter to the editor of the journal but I doubt they’d publish it, as I don’t speak the language of medical journals.

But, hey, let’s give it a try! I’ll go over to the webpage of Annals of Surgery, set up an account, write a letter . . .

Here it is:

Don’t calculate post-hoc power using observed estimate of effect size

Andrew Gelman

28 Mar 2018

In an article recently published in the Annals of Surgery, Bababekov et al. (2018) write: “as 80% power is difficult to achieve in surgical studies, we argue that the CONSORT and STROBE guidelines should be modified to include the disclosure of power—even if <80%—with the given sample size and effect size observed in that study.” This would be a bad idea. The problem is that the (estimated) effect size observed in a study is noisy, especially so in the sorts of studies discussed by the authors. Using estimated effect size can give a terrible estimate of power, and in many cases can lead to drastic overestimates of power (thus, extreme overconfidence of the sort that is rightly deplored by Bababekov et al. in their article), with the problem becoming even worse for studies that happen to achieve statistical significance. The problem is well known in the statistical and medical literatures; see, e.g., Lane and Dunlap (1978), Hedges (1984), Goodman and Berlin (1994), Senn (2002), and Lenth (2007). For some discussion of the systemic consequences of biased power calculations based on noisy estimates of effect size, see Button et al. (2013), and for an alternative approach to design and power analysis, see Gelman and Carlin (2014). That said, I agree with much of what Bababekov et al. (2018) say. I agree that the routine assumption of 80% power is a mistake, and that requirements of 80% power encourage researchers to exaggerate effect sizes in their experimental designs, to cheat in their analyses in order to attain the statistical significance that they was supposedly so nearly being assured (Gelman, 2017b). More generally, demands for near-certainty, along with the availability of statistical analysis tools that can yield statistical significance even in the absence of real effects (Simmons et al., 2011), have led to replication crisis and general corruption in many areas of science (Ioannidis, 2016), a problem which I believe is structural and persists even in the presence of honest intentions of many or most participants in the process (Gelman, 2017a). I appreciate the concerns of Bababekov et al. (2018) and I agree with their goals and general recommendations, including their conclusion that “we need to begin to convey the uncertainty associated with our studies so that patients and providers can be empowered to make appropriate decisions.” There is a just a problem with their recommendation to calculate power using observed effect sizes. References Button, K. S., Ioannidis, J. P. A., Mokrysz, C., Nosek, B., Flint, J., Robinson, E. S. J., and Munafo, M. R. (2013). Power failure: Why small sample size undermines the reliability of neuroscience. Nature Reviews Neuroscience 14, 1-12. Gelman, A. (2017a). Honesty and transparency are not enough. Chance 30 (1), 37-39. Gelman, A. (2017b). The “80% power” lie. Statistical Modeling, Causal Inference, and Social Science blog, 4 Dec. https://andrewgelman.com/2017/12/04/80-power-lie/ Gelman, A., and Carlin, J. B. (2014). Beyond power calculations: Assessing Type S (sign) and Type M (magnitude) errors. Perspectives on Psychological Science 9, 641-651. Goodman, S. N., and Berlin, J. A. (1994). The use of predicted confidence intervals when planning experiments and the misuse of power when interpreting results. Annals of Internal Medicine 121, 200-206. Hedges, L. V. (1984). Estimation of effect size under non- random sampling: The effects of censoring studies yielding statistically insignificant mean differences. Journal of Educational Statistics 9, 61-85. Ioannidis, J. (2016). Evidence-based medicine has been hijacked: a report to David Sackett. Journal of Clinical Epidemiology 73, 82-86. Lane, D. M., and Dunlap, W. P. (1978). Estimating effect size: Bias resulting from the significance criterion in editorial decisions. British Journal of Mathematical and Statistical Psychology 31, 107-112. Lenth, R. V. (2007). Statistical power calculations. Journal of Animal Science 85, E24-E29. Senn, S. J. (2002). Power is indeed irrelevant in interpreting completed studies. British Medical Journal 325, Article 1304. Simmons, J., Nelson, L., and Simonsohn, U. (2011). False- positive psychology: Undisclosed flexibility in data collection and analysis allow presenting anything as significant. Psychological Science 22, 1359–-366.

. . . upload it to the journal’s submission website. Done!

That took an hour. An hour worth spending? Who knows. I doubt the journal will accept the letter, but we’ll see. I assume their editorial review system is faster than this blog. Submission is on 28 Mar 2018, blog is scheduled for posting 24 Sept 2018.

The post Don’t calculate post-hoc power using observed estimate of effect size appeared first on Statistical Modeling, Causal Inference, and Social Science.

Author

发表评论

电子邮件地址不会被公开。 必填项已用*标注