[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
Diego Bellavia <messadua@yahoo.it> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: RE: Effect size and Power for the difference between means |

Date |
Fri, 17 Aug 2007 22:03:48 +0000 (GMT) |

Hey Tom, I really like your conclusions, stats is math, and creative use of tools improves the options and STATA itself. you have a nice weekend too Diego ----- Messaggio originale ----- Da: "Steichen, Thomas J." <SteichT@rjrt.com> A: statalist@hsphsun2.harvard.edu Inviato: Venerdì 17 agosto 2007, 16:08:14 Oggetto: RE: st: RE: Effect size and Power for the difference between means Hi Diego, I thought from your message that you expected to get a power or sample size from -sizefx- program, and I had hoped to communicate that -sizefx-, which only provides effect sizes from raw data, is not directly useful for computing power, etc. That is, -sizefx- alone will not provide a power computation. You are, of course, correct that having an estimated effect size is a crucial element in computing power. I was also perhaps overly negative about -sizefx- because I have predominantly used effect sizes in meta-analyses of results from literature, so I'm used to thinking about Cohen's d, etc., as being computed from summary statistics rather than raw data. Thus a program that does so from raw data seemed a somewhat useless exercise; I had forgotten that one might directly insert an ES in a power calculation rather than using means, sd's, etc. This discussion, like many on Statalist, is a reminder that many statistical tools have multiple uses; and my use might not be your use. Have a good weekend, Tom ----------------------------------- Thomas J. Steichen steicht@rjrt.com ----------------------------------- -----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Diego Bellavia Sent: Friday, August 17, 2007 4:24 PM To: statalist@hsphsun2.harvard.edu Subject: Re: st: RE: Effect size and Power for the difference between means Tom, Thank you for the clear and thorough explanation on effect size. I agree with all you sayd and now I understand that my question was a non sense. But, I do not understand why ES is not useful for sample size determination in power analysis ? As you pointed out, in its very basic formulation, ES reports the number of SDs (d) the outcome variable is increased (o decreased) in the treatment group compared to the control group. Therefore, knowing ES you have an immediate estimation of the impact the specific treatment has in your study population, and being a ratio of delta (m1 - m2) / pooled SD (outcome variable) you take in account variability of the outcome variable a the same time. All the commands I reviewed in STATA to compute sample size (given power) or power (when sample size is already known) use the difference in the means (m1 - m2) and standard deviation of the two groups (sampsi command) or directly the ES (fpower). However you prefer to put the things, separating the difference of the means in the two groups and the SD of the outcome variable or putting them together in a ratio (that is ES), ES remains a critical measure to get sample size (or power) in a power analysis. All this, naturally, IMHO. If I lost something here, please help me to learn. Thank you Diego ----- Messaggio originale ----- Da: "Steichen, Thomas J." <SteichT@rjrt.com> A: statalist@hsphsun2.harvard.edu Inviato: Giovedì 16 agosto 2007, 17:32:53 Oggetto: st: RE: Effect size and Power for the difference between means Diego, My impression, given what you say in your message, is that you may have a misunderstanding of what -sizefx- is intended to compute. Generally, an effect size is a mean difference divided by the standard error of that difference (or, alternatively, is the correlation of two variables divided by the standard error of that correlation). This has nothing to do with computing sample size, power, etc. -sizefx age1 age2- merely computes the mean, variance and sample size for each variable (age1, age2) then computes the various effect sizes. For example, Cohen's d is computed as: d = abs(m1-m2) / sqrt(((n1*v1)+(n2*v2))/(n1+n2)) where m1 is the mean, n1 is the sample size and v1 is the variance of age1 and m2 is the mean, n2 is the sample size and v2 is the variance of age2 d is then read to indicate that the means of age1 and age2 differ by d standard deviations. Clearly, you do not calculate the variables age1 or age2, you just use them in the program. It is probably fair to note that the typical use of effect sizes is when t-values, correlations or the sufficient statistics are all you have (i.e., you do not have raw data). Then you just directly plug in what you have into an appropriate formula and spit out the effect size. Because of this, I do not see -sizefx- as a particularly useful program. Tom ----------------------------------- Thomas J. Steichen steicht@rjrt.com ----------------------------------- -----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Diego Bellavia Sent: Wednesday, August 15, 2007 10:47 AM To: STATAlist Subject: st: Effect size and Power for the difference between means Dear Statalisters, I am sorry for the basic questions, but I have a problem trying to figure out how the sizefx command works in STATA 9.2. 1) Let us say I have two groups and some outcome variables for which I would like to calculate the ES. Looking at the help file of sizefx, examples are like sizefx age1 age2 sizefx before after (matched design) The problem is that I do not know how to calculate age1 and age2 (I know this is very very basic, please mercy on me). Then I tryed something like sizefx groups age to see the ES of age in the different groups, and I got something like Effect Size Measures: Cohen's d and Hedges' g ----------------------------------------------------- Cohen's d statistic (pooled variance) = 2.2798227 Cohen's d statistic (t-statistic method) = 2.2847456 Hedges' g statistic = 2.2847456 Effect size correlations: r ----------------------------------------------------- ES correlation (Cohen's d method) r = .75173368 ES correlation (t-statistic method) r = .75214284 ES correlation (Hedges' g method) r = .75214284 No problems with the r but regarding the Cohen's d and the Hedge's g 2.27 and 2.28 values become 0.227 and 0.228 respectively ? Another weird thing is that sometimes I have higher ES for variables that are not significantly different between the two groups...how is it possibile ? 2) Regarding the sample size calculations for differences between means what command do you reccomend ? I found "sampsi" (t-test) then "fpower" and "simpower" (for ANOVA). Is this a complete list or do you use something else ? Are there improvements in STATA 10 regarding this ? Thank you in advance for the help Diego ___________________________________ L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ ----------------------------------------- CONFIDENTIALITY NOTE: This e-mail message, including any attachment(s), contains information that may be confidential, protected by the attorney-client or other legal privileges, and/or proprietary non-public information. If you are not an intended recipient of this message or an authorized assistant to an intended recipient, please notify the sender by replying to this message and then delete it from your system. Use, dissemination, distribution, or reproduction of this message and/or any of its attachments (if any) by unintended recipients is not authorized and may be unlawful. * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ ___________________________________ L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ ___________________________________ L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**Re: st: dates** - Next by Date:
**st: Goodness-of-fit before MonteCarlo simulation** - Previous by thread:
**Re: st: RE: Effect size and Power for the difference between means** - Next by thread:
**st: Goodness-of-fit before MonteCarlo simulation** - Index(es):

© Copyright 1996–2017 StataCorp LLC | Terms of use | Privacy | Contact us | What's new | Site index |