On-line statistics

Key for the selection of a test for prediction

Regression analysis generally does not follow the normal mode of sampling i.e. extracting individuals from a population at random and then measuring a variable or two. Instead we choose points in time or position along a river or hedgerow or concentrations of toxin. Therefore we measure the response at each point and do not select at random.

As with correlations, get into the habit of drawing a scatter graph of your data before any analyses are carried out. You should be able to judge whether a relationship exists from such a simple act.

Use the key by answering the questions in the most relevant way. It is advised that you double check through a good reference guide.

1. Have you got binomial data?

Yes...........go to Binomial distribution and predictions

No............go to 2

2. Is the relationship between the data sets linear?

Yes...........go to 3

No............try transforming, otherwise use non-linear regression (not covered by these pages)

3. Have you chosen the values of the independent variable?

No...........type-two regression (see Zar or Sokal and Rohlf)

Yes..........go to 4 (type-one regression)

4. Have you got one or more than one set of x -values?

One..................go to 5

Two or more.....multiple regression

5. After performing the analysis do your residual plots show an even spread across the graph and does the normal probability plot follow a linear line? (refers to equal variances between, and normal distribution of y -values)

No......Kendall's non-parametric regression (see also Zar or Sokal and Rohlf)

Yes......simple linear regression

---

Descriptive Stats

Diversity Indices

Comparisons

Correlations

Regression