The FastDay Forum

[carorees]

Hi Moogie

I was just wondering if it would be a simple matter to extract the range as well as the median for the forum wide stats? And/or calculate the standard deviations to go with the means? It would be helpful for our research into why some people lose weight faster than others.

I don't want to create a lot of work for you, but if it is easy perhaps you could add it for us?

[Moogie]

I'll have to look into what those are before I can tell you how easy they'd be to add

Is range just the minimum & maximum (eg top and bottom end of the data, where median is in the middle)?

SD... I vaguely recall it from doing stats at school, but that was a long time ago now. I'll need a proper refresher on what it is and how to calculate it!

At the moment my priority is the recipes section to make the site a bit more comprehensive but there are a couple of things I want to add to the tracker at some point so can try to do the SD/range then if possible!

[Moogie]

Haha, I feel like a kid.
http://www.mathsisfun.com/data/standard-deviation.html

Looks like I have a formula to get the SD.... does it need to be graphed somehow or do you just need the figure?

[carorees]

Range is just max and min as you say. Glad you have a formula for SD, my calculator does it at the touch of a button so I've forgotten the formula too! Nothing fancy needed (unless Dominic or Phil think otherwise!) just the figures. That would be marvellous!

Standard deviation is just a number, but of course a different number for each sample set. You would derive it from the population variance calculation for each sample set BTW (because for each sample you have all the data) - not the sample variance.

I think the range is as you thought just the difference between the highest and lowest values - not very meaningful especially as these values may well be mistakes. However the standard deviation does have value because it shows the variability of the dataset and also allows some independent confidence in the statistical conclusions. A relatively large standard deviation either means there is a lot of natural variation or that the data is unreliable (e.g. that there are a lot of mistakes or just that there isn't enough data to make it useful statistically).

[carorees]

I think it would help us to choose cut offs for the survey so rather than dividing the population into quartiles we could look at those more than n SDs from the mean. Also it would help with advising whether people are experiencing the expected weight loss.

Scarry, we had to prove the Standard Deviation formula in Maths A Level, I could never remember how to do it and ended up getting an E grade, I still became an accountant mind!

Mean and SD is mainly used for normally distributed data (symmetrical Gauss curve on a histogram). Weight is often not normally distributed, with a steep slope on the left side, and a slow taper on the right side of the curve. You can see the difference between mean and median in the progress tracker indicating weight loss per subgroup. which suggests that the weight loss data are not normally distributed)
Range can be interesting, but a single outlier influences it tremendously. Interquartile range (25-75%) is often used, but you could also use 10-90 percentiles.
For SD, values between -2 and +2SD would tell you where p95% of the population lies. -1.3 to +1.3 SD describes approximately the middle 80% of the population. Again, assuming a symmetrical distribution.

It all depends on what you want to do with it.

[Moogie]

I'm afraid most of that is gobbledegook to me but I think an SD value should be fairly simple to add for any given selection on the tracker.

[Moogie]

Just posting a couple of links here for me to refer to later...
http://php.net/manual/en/function.stats ... iation.php
(will have to check if the stats stuff is installed on the server)

http://www.devshed.com/c/a/PHP/Performi ... ith-PHP/3/

Excel can do a standard deviation value very easily. It can also do some interesting statistical tests now, including ANOVAs, but you have to have a normal distribution of variance (symmetrical curve on either side of the mean).

[Moogie]

Thanks DrLCH Unfortunately I don't think it's possible to install Excel on the server and tie it in with the PHP & database for the site - but perhaps it's worth me making the stats downloadable for others to analyse? (with userids replaced to preserve anonymity of course)

will have to check if the stats stuff is installed on the server

If not, the user-contributed notes on the page of your first link provide ways to calculate the SD without needing it (slower I expect).