Statistics Flashcards
What analysis should you use of there is 2 categorical columns
odds ratio, g test or chi squared test
What analysis should you use if there is one categorical column and 1 normal column
compare groups (groups in the categorical column)
What analysis should you use if there are 2 normal columns
straight line, correlation OR repeated measures
How can you choose between using straight line or correlation or repeated measures
You use a straight line or correlation when:
- The two columns seem to be very different kinds of things (age, blood pressure).
- When you might want to predict one column based on the values in the other (same).
- When you think these two different things might be related.
You use repeated measures when
* The two columns seem to be very similar kinds of things (stopping distances in wet, dry)
* When you’re interested in whether they are the same in some way.
how can we tell if there’s any extreme values before using mean and median
use a histogram or boxplot of the data to see if there are any extreme values
boxplot
The median (50% of the data is below the median, 50% above)
* The first quartile (25% of the data is below the 1st quartile value)
* The third quartile (75% of the data is below the 3rd quartile value)
* The Interquartile range (IQR) = 3rd quartile value – 1st quartile value.
based on the box plot when should the mean and medians be used
mean can be used:
when the median line is in the middle of the box
the whiskers aren’t too long
there are no plotted points outside the whiskers
And also:
mean is similar to median
paired measurements
occur when we take two or more measurements on a number of people. can analyse by looking at the distance between the two.
what plot do we use for compare two groups
box plot
what plot do we use for paired measurements
bland altman plot
bland altman plot
horizontal black line is median of Y values
red line is at zero. if theres no difference the data should cluster around this
grey area should cover 95% of data points
how to calculate odds ratio
probability event will occur over probability it wont
P/(1-p)
analysing odds ratio
odd ratio of 1 indicates no relationship between rows and columns
odds ratio only works on tables with four cells