Hi Helen, Brian saved me some typing. One more thing: power calcs are really quick and easy.* Again, that's a nice correction, if you get it. Jeremy *Even though people who do them for a job (like me) pretend they're not. On 22 October 2010 09:58, Brian K. Saxby <[log in to unmask]> wrote: > Hi Helen, > > I agree, try to relax and enjoy the attention if you can! > > The power issue is important during the planning stages of a study, > especially in clinical trials (the area I work in), where there's a very > specific hypothesis to be tested and a drug may achieve or fail to gain > approval for market based on the outcome of a study. Even in these trials > however, it's often only the primary outcomes measures that are adequately > powered, and the secondary measures provide supportive information but may > not necessarily reach statistical significance. Although the p value is > important, in terms of meaningful results the effect size is also very > relevant - it's possible to see an effect that doesn't reach p<0.5 but is > close, and if the effect size is meaningful, then the assumption is that it > would've got there if more subjects were tested. > > It is possible to calculate power retrospectively, but personally I only do > it when I'm trying to determine whether a 'close' result was powered or not > (i.e. trying to distinguish between a near-miss versus an adequately-powered > fail). If you have some results that are borderline, it may be worth > checking to see if you were powered or not in case it comes up at viva, and > you could go as far as to work out how many subjects you would need to have > sampled to detect a statistically significant difference (assuming the > effect size you've seen is the same as what exists in the rest of the > population) - but you can do this just on your own data - I'm not sure at > this stage though what you'd get from trawling through the papers you got > your scales from - it may have helped with powering the study at the start, > but your own data are likely to be more informative given that they're from > using the scales with the subjects you're interested in. I wouldn't get too > hung up on it though - even if it does come up (and remember Jeremy's advice > that not many PhD examiners like to go into stats details themselves!), > having an awareness of the issues of power and effect size is probably > enough, and worst-case scenario would be you have to put in some > retrospective power calcs as corrections. > > Good luck! > > Brian > > > > On 22/Oct/2010 07:32, Jamison-Powell, Susan wrote: > > Hi Helen, > > > > I am currently preparing my thesis for submission and am teaching on a > Master’s advanced stats course. Guess what we did last week? Yes, that’s > right - power. I was having the same concerns as you. I have decided to > comment upon the power in the results/discussion sections when discussing > null results. I was very much tied to opportunity samples (and longitudinal > ones – so attrition is something I think I will be quizzed on as my > attrition rates are awful), so like you I was limited to the number of > people I could persuade to take part. > > > > I have been told that your viva examiners never ask you the questions you > think they will, so I would take a deep breath, concentrate on getting deep > and intimate with your thesis and think of your viva as an opportunity to > exclusively talk about your research for a couple of hours (because I doubt > we will get that chance again!). > > > > Let us know how you get on. > > > > Sue > > > > From: Research of postgraduate psychologists. > [mailto:[log in to unmask]] On Behalf Of Helen Mann > Sent: 22 October 2010 12:26 > To: [log in to unmask] > Subject: Re: Power calculations > > > > Hello (yes me again, im revising for my viva next week and getting paranoid) > > I never did any sample size power calculations before I ran my > test....mainly because I never knew about them at the time and thought you > just went with how many people you could get to take part in your > experiment.....anyway, do you think I need to go through all the papers > where I got my scales from and work out the sample size I needed > for completing each scale to get power = 0.8???!! > > Or can I use my data to work out the sample size power and then say whether > I needed more or less participants?!!? > > Has anyone ever been asked about sample size in a viva??!?! > > HELP!! > > Helen -- Jeremy Miles Psychology Research Methods Wiki: www.researchmethodsinpsychology.com