(Y2) Statistical testing

• A statistical test is used to determine whether a difference or association/correlation found in a particular investigation is statistically significant (i.e. whether the result could have occurred by chance or there is a real effect)
• Three criteria to choose a statistical test:

  1. Looking for a difference or a correlation/association
  2. Is experimental design related (repeated measures/matched pairs) or unrelated (independent groups)
  3. What is the level of measurement?
• When can Chi-Squared be used?

  Nominal data, test of difference- related design

  Nominal data, test of correlation or association
• When can Mann-Whitney be used?

  Ordinal data, test of difference- unrelated design
• When can the Sign test be used?

  Nominal data, test of difference- related design
• When can the Wilcoxon test be used?

  Ordinal data, test of difference- related design
• When can Spearman's Rho be used?

  Ordinal data, test of association or correlation
• When is the unrelated t-test used?

  Interval data, test of difference- unrelated design
• When is the related t-test used?

  Interval data, test of difference- related design
• When is Pearson's r used?

  Interval data, test of association or correlation
• Nominal data- categories:

  Each item can only appear in one category. There is no order.

  For example, people naming their favourite football team
• Ordinal data- placed in order, intervals are subjective:

  Data is collected on a numerical, ordered scale but intervals are variable, so that a score of 8 is not twice as much as a score of 4

  Ordinal data lacks precision because it is based on subjective opinion rather than objective measures

  For example, asking someone to rate how much they like psychology on a scale of 1 to 10
• Interval data- units of equal size:

  Interval data is based on numerical scale that include units of equal, precisely defined size

  This includes counting observations in an observational study (8 tallies is twice as much as 4 tallies) or any 'public' unit of measurement (e.g. time, temperature, length)

  Interval data is 'better' than ordinal data because more detail is preserved as the score are not converted to ranks