Reliability analysis
Analyze scale reliability analysis
Click statistics and tick boxes for ‘item’, ‘scale’ and ‘scale if deleted’ in descriptive and
‘correlations’ in inter-item.
Recode items
look at the Inter-Item Correlation Matrix in the output of our reliability analysis.
Are all items positively correlated? If yes, then this suggests that they are all formulated in
the same direction. If an item is negatively correlated with the rest of the items, this
suggests that this item is formulated in the reversed direction and you need to recode it.
Transform recode into different variable old and new values take the value and give
it the reverse.
Possible questions
“Do all items contribute positively to the reliability of the scale?”
look at the corrected item-total correlations: are the positive and above .3. also look at
the Cronbach’s alpha if item is deleted and that should be lower than the alpha of the whole
scale.
Deleting items from the scale will lower reliability.
“Which item contributes the most to the reliability of the scale?”
the highest corrected item-total correlation and the lowest Cronbach’s alpha if item
deleted.
“Which items are most related?”
look at the correlations between the items highest squared multiple correlation and
high corrected item-total correlation.
Compute total scores (sum scores)
Transform compute put in first variable + second etc.
Syntax:
COMPUTE SumScores=SUM(v266, v267, v268, v269, v270, v271, v272, v273, v274, v275).
EXECUTE.
Conduct a principal component analysis (PCA)
Analyze dimension reduction factor
Click ‘KMO and Bartlett’s test of sphericity’ in the Descriptives menu + ‘the method principal
components’ + ‘Scree plot’ in the Extraction menu + ‘sorted by size’ and ‘suppress small
coefficients’ in the options menu (absolute value below .30)
Conditions to run a PCA
- KMO > 0.6
- Bartlett is significant (p-value < 0.05)
The communality of an item is the amount of variance in that item that is explained by all
components. It is a measure of how well the components explain people’s answers to that
item can be found in the table ‘commualities’
Analyze scale reliability analysis
Click statistics and tick boxes for ‘item’, ‘scale’ and ‘scale if deleted’ in descriptive and
‘correlations’ in inter-item.
Recode items
look at the Inter-Item Correlation Matrix in the output of our reliability analysis.
Are all items positively correlated? If yes, then this suggests that they are all formulated in
the same direction. If an item is negatively correlated with the rest of the items, this
suggests that this item is formulated in the reversed direction and you need to recode it.
Transform recode into different variable old and new values take the value and give
it the reverse.
Possible questions
“Do all items contribute positively to the reliability of the scale?”
look at the corrected item-total correlations: are the positive and above .3. also look at
the Cronbach’s alpha if item is deleted and that should be lower than the alpha of the whole
scale.
Deleting items from the scale will lower reliability.
“Which item contributes the most to the reliability of the scale?”
the highest corrected item-total correlation and the lowest Cronbach’s alpha if item
deleted.
“Which items are most related?”
look at the correlations between the items highest squared multiple correlation and
high corrected item-total correlation.
Compute total scores (sum scores)
Transform compute put in first variable + second etc.
Syntax:
COMPUTE SumScores=SUM(v266, v267, v268, v269, v270, v271, v272, v273, v274, v275).
EXECUTE.
Conduct a principal component analysis (PCA)
Analyze dimension reduction factor
Click ‘KMO and Bartlett’s test of sphericity’ in the Descriptives menu + ‘the method principal
components’ + ‘Scree plot’ in the Extraction menu + ‘sorted by size’ and ‘suppress small
coefficients’ in the options menu (absolute value below .30)
Conditions to run a PCA
- KMO > 0.6
- Bartlett is significant (p-value < 0.05)
The communality of an item is the amount of variance in that item that is explained by all
components. It is a measure of how well the components explain people’s answers to that
item can be found in the table ‘commualities’
