statistical tests

• What are parametric stats?

  • More powerful than non-parametric tests
  • Use raw data in their calculations and take into account all the information available
  • Make calcs using mean and standard deviation
• why are parametric stats stronger than non-parametric stats?

  • Non parametric use only ranked data- lose some of the detail
  • Greater power efficiency
  • Require less data in order for us to reject null hypothesis
  • Can detect significance where non parametric tests don’t
• what 3 criteria must the data satisfy in order to use a parametric test?

  • data must be interval or ratio measurement
  • scores must be drawn from a normally distributed population
  • variances must not be too different- the square of the standard deviation in each sample must be more or less the same/similar
• what does < > mean in terms of probability?

  • < = less than (p<0.5)
  • > = greater than (p>0.8)
• what does lenient and stringent levels mean in terms of significance?

  • lenient- easy to achieve, allowing a greater probability that results are not due to chance
  • stringent- hard to achieve, not allowing a great probability that results are not due to chance
• what is a type I error?

  • psychologist/person testing for difference claims that a difference between data sets is real when there's in fact no real difference/significance because the level of significance is too high/lenient
  • so researcher rejects null hypothesis when it should be accepted
  • false positive
• what is a type II error?

  • a real/significant difference is overlooked because the level of significance is too low/stringent and harder to achieve
  • so you accept a null hypothesis when it should be rejected
  • false negative
• what 4 things must be considered when deciding what statistical test to carry out?

  1. is it a test of difference or an association?
  2. what research design should be used? (matched paris, rep measures)
  3. is it nominal/ordinal/interval?
  4. If it's interval does it fit the requirements for a parametric test?
• in what tests does the calculated value have to be less than/equal to the critical one to detect significance?

  • Mann-Whitney U
  • Wilcoxon
  • sign test
• in what tests does the calculated value have to be greater than/equal to the critical one to detect significance?

  • related T test
  • unrelated T test
  • Pearson's R
  • Chi squared
  • Spearman's Rho
• what things must be checked when looking for a critical value?

  • the hypothesis is in the right direction
  • you have the correct degrees of freedom (number of values for each data set)
  • find the right level of significance (0.05 or 5%)
  • determine which has to be greater for significance- calculated/critical. IGNORE ANY SIGNS (+ve/-ve numbers) WHEN IDENTIFYING WHICH OUT OF THE CALCULATED/CRITICAL VALUE IS LESS/GREATER
• when are bar charts used and when are histograms used?

  BAR CHARTS-

  • when the data is discontinuous
  • when the data is represented as distinct categories/2 different conditions
  • has no true zero
  • space left in-between bars to indicate lack of continuity

  HISTOGRAMS-

  • when the data is continuous
  • when the data is of numerical value/collected quantitively
  • there is a true zero
  • bars are proportional to frequencies represented
  • no space between bars to indicate continuity
• what is meant by continuous and discontinuous data?

  • continuous- numerical value, quantitative, has a particular order
  • discontinuous- data is collected using categories/2 distinct conditions, has no particular order
• recite the statement used for declaring or refusing significance

  the results are/are not significant, as the critical value of... (number stated) is less than the calculated value of (number stated) at the 0.05/5% level (df=....) therefore we can accept/reject the null hypothesis and conclude there's less/more than a 5% probability that the results are due to chance
• why do we use the 5% level in psychology?

  strikes a balance between the risk of making a type one or a type two error
• when should a non-directional hypothesis be used?

  • when there's no previous research
  • when previous research is contradictory