Inferential Statistics

What are significance tests for?

• Significance tests are used to see just how far an IV caused a DV in an experiment, as opposed to it being due to chance
• They work on probability, with 1 = absolutely will happen, 0 = absolutely won't happen

Hypotheses

• Experimental - the IV affected the DV, and a significant difference was seen
• Null - the IV had no effect on the DV, and any difference was due to chance

Which level of significance should we use?

• <50% or <0.5 - NO
• This would give too much margin for error, leading us to reject the null when we had really ought to accept it
• The results were due to chance after all
• This is called a Type 1 error

• <1% or <0.01 - NO
• This is too strict: we might accept the null when we should have rejected it
• The results were due to the IV after all
• This is called a Type 2 error

• <5% or <0.05 - YES
• This is ideal for social science
• A 5% probability that it was due to chance - we can live with

Nominal

• When data is put into categories or just named
• E.g. favourite subject at school, did you finish the race (yes/no)

Ordinal

• When data is ranked
• You can line them up but the difference between them might not be the same
• E.g. positions in a race, strongly agree/disagree

Interval

• Has equal measurements on a measurement scale, e.g. temperature: there is the same difference between 4 and 6 degrees as between 10 and 12 degrees
• No true zero - zero doesn't mean nothing. 0 degrees doesn't mean no temperature and 0 on an intelligence test doesn't mean you have no intelligence

Ratio

• Has a true zero, e.g. weight and distance: 0 kgs really is no weight

Is it nominal data?

• Yes - use CHI-SQUARE
• No - move on

Is it a correlation?

• Yes - use SPEARMAN'S RHO
• No - move on

Is it a test of difference between two sets of scores?

• Yes - use WILCOXON T or MANN-WHITNEY U

Which one do you use?

• Repeated Measures - WILCOXON T
• Independent Measures - MANN-WHITNEY U