Psychometrics SPSS Workgroup Answers
Week 1
'Calculate scale scores by finding the mean of the items concerned'
- First, check in the Annex which items you need.
- Then, compute a new variable: (Transform > Compute Variable) and apply the
function MEAN.n.
'What does a high score mean?'
- Check the Annex for an explanation of the scoring methods.
Create a report/Provide relevant statistics for each of the scale scores across the total
research group.
- Analyze > Descriptive Statistics > Descriptives, and request mean, standard
deviation, minimum and maximum.
Produce graphs of the distribution of the scale scores > describe these distributions.
Histogram: this is a distribution of the means (scale scores).
What should I describe?
Example = right skewed, which makes sense given the population, we would expect more
kids to have a pleasant relationship than a neglectful relationship, meaning a higher mean.
Boxplot: check 'simple' and 'summaries of separate variables'. Put the scale scores in 'boxes
represent'.
What should I describe?
Example = a lot of outliers, also pointing at a skewed result.
Calculate z-scores.
- Analyze > Descriptive Statistics > Descriptives, add relevant variables (the new
computed variable which is the scale score) and select 'save standardized values as
variables'.
Do the values look reasonable?
For z-scores, look if the z-scores that are closer to 0 are closer to the mean in the raw data.
Calculate T-scores.
- Compute a new variable (Transform > Compute Variable).
- Fill in T= z*10+50.
Do the values look reasonable?
Eyeball it, is it around 50 when z=0 or 50 higher in general.
, Get SPSS to determine percentile scores.
Transform > Rank cases; via Rank Types uncheck 'Rank' and select 'Fractional rank as %';
via Ties select 'High'. (It does not matter which variable you enter here (raw, z or T), since
they all should be distributed the same, they're just translations of one another).
Do the values look reasonable? Are the scores between 0 and 100.
Round off percentages to whole numbers by setting the number of decimal places of the
variable to 0 in Variable View.
Create a norm table.
Do the scales separately, so you create two tables!
- Analyze > Reports > Case Summaries. Put the normed scores (z, T, percentile
scores) in the Variables box, and the scale score (the mean) (one at a time
(separately for each group)) in the Grouping Variable(s) box. Uncheck 'Display
cases'. Via Statistics ask for cell means only (so not 'number of cases').
Check whether the norm table is 'correct' by looking at minima and maxima of the
various normed score distributions.
When compared to the earlier results, it should look similar. So if it was skewed to the right
earlier, it should still show those same results.
Comparisons between the two groups can be made as well.
Week 2
Split-Half
Split the items of a scale into two parts that are equivalent with respect to certain item
characteristics.
Only LOKFA 1 - 7, 9 - 11 (because of arbitrary rules that 6 items are not enough, so the
other scale will not be used).
- Analyze > Descriptives > Plug in LOKFA above & click on 'options' and then
'descending means'; 'mean' & 'standard deviation' > OK.
- Analyze > Scale > Reliability analysis
Group 1 Group 2
5 6
1 2
4 9
Week 1
'Calculate scale scores by finding the mean of the items concerned'
- First, check in the Annex which items you need.
- Then, compute a new variable: (Transform > Compute Variable) and apply the
function MEAN.n.
'What does a high score mean?'
- Check the Annex for an explanation of the scoring methods.
Create a report/Provide relevant statistics for each of the scale scores across the total
research group.
- Analyze > Descriptive Statistics > Descriptives, and request mean, standard
deviation, minimum and maximum.
Produce graphs of the distribution of the scale scores > describe these distributions.
Histogram: this is a distribution of the means (scale scores).
What should I describe?
Example = right skewed, which makes sense given the population, we would expect more
kids to have a pleasant relationship than a neglectful relationship, meaning a higher mean.
Boxplot: check 'simple' and 'summaries of separate variables'. Put the scale scores in 'boxes
represent'.
What should I describe?
Example = a lot of outliers, also pointing at a skewed result.
Calculate z-scores.
- Analyze > Descriptive Statistics > Descriptives, add relevant variables (the new
computed variable which is the scale score) and select 'save standardized values as
variables'.
Do the values look reasonable?
For z-scores, look if the z-scores that are closer to 0 are closer to the mean in the raw data.
Calculate T-scores.
- Compute a new variable (Transform > Compute Variable).
- Fill in T= z*10+50.
Do the values look reasonable?
Eyeball it, is it around 50 when z=0 or 50 higher in general.
, Get SPSS to determine percentile scores.
Transform > Rank cases; via Rank Types uncheck 'Rank' and select 'Fractional rank as %';
via Ties select 'High'. (It does not matter which variable you enter here (raw, z or T), since
they all should be distributed the same, they're just translations of one another).
Do the values look reasonable? Are the scores between 0 and 100.
Round off percentages to whole numbers by setting the number of decimal places of the
variable to 0 in Variable View.
Create a norm table.
Do the scales separately, so you create two tables!
- Analyze > Reports > Case Summaries. Put the normed scores (z, T, percentile
scores) in the Variables box, and the scale score (the mean) (one at a time
(separately for each group)) in the Grouping Variable(s) box. Uncheck 'Display
cases'. Via Statistics ask for cell means only (so not 'number of cases').
Check whether the norm table is 'correct' by looking at minima and maxima of the
various normed score distributions.
When compared to the earlier results, it should look similar. So if it was skewed to the right
earlier, it should still show those same results.
Comparisons between the two groups can be made as well.
Week 2
Split-Half
Split the items of a scale into two parts that are equivalent with respect to certain item
characteristics.
Only LOKFA 1 - 7, 9 - 11 (because of arbitrary rules that 6 items are not enough, so the
other scale will not be used).
- Analyze > Descriptives > Plug in LOKFA above & click on 'options' and then
'descending means'; 'mean' & 'standard deviation' > OK.
- Analyze > Scale > Reliability analysis
Group 1 Group 2
5 6
1 2
4 9
