Statistics: The basics

 Posts: 26
 Joined: Sun Mar 25, 2007 6:43 pm
 Contact:
Statistics: The basics
For a basic overview of statistical principles, go to:
http://www.statisticshell.com/html/apf.html
Levels of data I remember by the french word for black; NOIR
Nominal  Categories like male / female
Ordinal  Numerical things that can be ranked but not measured e.g. Likert 15 scale (disagree > agree strongly)
Interval  Numerical things with an even gap in between them but with an arbitrary zero e.g. time, height above sea level, longitude
Ratio  Numerical things with an even gap with an absolute zero, e.g. weight, height, temperature. Also sometimes called continuous data.
Parametric tests like the TTest have a set of assumptions that go with them. For instance the data you use must normally be at the very least ordinal level data, preferably interval. These tests assume that the data you're looking at is at least similar to the normal distribution (that good old bell curve you know and love).
If the data is not normally distributed (number of legs in the population is a nicely skewed one!) or the level of data is only categorical than you're looking at a nonparametric test. You might also need to use one of these if the data is highly skewed or if your standard deviation is very wide. As a rule of thumb if your standard deviation is almost as big as your mean then you've probably not got a normally distributed set of data. You should also use a nonparametric test when you have a small sample size; this will vary depending on what you're measuring but you'd be hardpressed to justify a parametric test with a sample size of less than 1520 per group.
That said some tests like ANOVA are quite robust despite data not conforming to the normal distribution. Then it all starts becoming a little fuzzier...
When reporting on posthoc test differences include the magnitude of the difference e.g. "A posthoc Bonferronni test revealed that sporadic ALS patients performed more poorly than control participants (mean difference = 2.6 points, 95% CI = 0.61  4.65, p=0.017)."
There are lots of short primers on stats on the net. So well worth a Google if you get beyond the Andy Field stuff.
Last checked by qualified clinical psychologist (BlueCat) on 23/05/2016
http://www.statisticshell.com/html/apf.html
Levels of data I remember by the french word for black; NOIR
Nominal  Categories like male / female
Ordinal  Numerical things that can be ranked but not measured e.g. Likert 15 scale (disagree > agree strongly)
Interval  Numerical things with an even gap in between them but with an arbitrary zero e.g. time, height above sea level, longitude
Ratio  Numerical things with an even gap with an absolute zero, e.g. weight, height, temperature. Also sometimes called continuous data.
Parametric tests like the TTest have a set of assumptions that go with them. For instance the data you use must normally be at the very least ordinal level data, preferably interval. These tests assume that the data you're looking at is at least similar to the normal distribution (that good old bell curve you know and love).
If the data is not normally distributed (number of legs in the population is a nicely skewed one!) or the level of data is only categorical than you're looking at a nonparametric test. You might also need to use one of these if the data is highly skewed or if your standard deviation is very wide. As a rule of thumb if your standard deviation is almost as big as your mean then you've probably not got a normally distributed set of data. You should also use a nonparametric test when you have a small sample size; this will vary depending on what you're measuring but you'd be hardpressed to justify a parametric test with a sample size of less than 1520 per group.
That said some tests like ANOVA are quite robust despite data not conforming to the normal distribution. Then it all starts becoming a little fuzzier...
When reporting on posthoc test differences include the magnitude of the difference e.g. "A posthoc Bonferronni test revealed that sporadic ALS patients performed more poorly than control participants (mean difference = 2.6 points, 95% CI = 0.61  4.65, p=0.017)."
There are lots of short primers on stats on the net. So well worth a Google if you get beyond the Andy Field stuff.
Last checked by qualified clinical psychologist (BlueCat) on 23/05/2016
Who is online
Users browsing this forum: No registered users and 1 guest