Embedded Files
General Purpose
This is often known as Cultural Consensus Analysis and is best seen with an example.
You have 9 people who are asked the same set of 20 questions. Based on their answers, you want to know if they form a relatively homogeneous group. Consensus analysis determines if there is one group, two groups or no group.
There are a few constraints. Perhaps the most important is that all the questions must be of the same type (e.g., of equal weight and character). Also, everyone needs to answer all the questions. The questions are generally T/F or multiple choice.
You can generalize this example by assuming that "people" could also mean "places" or even "things." The "questions" might be different kinds of "things" or "places."
Running and Interpreting the Analysis
Prepare the data either in PAST, a spreadsheet (e.g., Excel) or as a comma delimited file (CSV). See the example on the right.
Start PAST and enter the data (use the Edit checkboxes to allow entry and placement of headings). Uncheck the boxes when the data are ready for analysis.
- Highlight the data area.
- Click on the Multivar menu and choose Principal Components.
- The analysis box that appears should be changed by clicking on the Correlations radio button.
- First you want to see if the responses have an overall similarity. Use the first two Eigenvalues that are listed. Divide the first one by the second one. If this number is equal to or greater than 3, you have a consensus (and therefore all the respondents are acting like one relatively homogeneous group).
- It is good to check the Loadings (click on the View Loadings button). When the Loadings window appears, make sure to click the Correlations radio button. Positive loading (bars are up) indicate group membership. Long positive bars show that a person is highly representative of the overall group.
- The second test is used if the first one fails (i.e., the value of the ratio of the first and second eigenvalues is less than three). This test is the ratio of the second and third eigenvalues; this needs to be equal to or greater than 3 to indicate that the individuals can be divided into two groups.
- Click on the View Loadings and click on the Correlations radio button. You should see grouping among the bars indicating the division of the individuals into the two groups (assuming that you passed the >= 3 test described above). If this is the case, you would then use your demographic (or other data) to explain the group membership.
The analysis to the right shows the interpretation when it involves two groups.
Example Data
Binary (T/F) data are used in this questionnaire administered to nine people.
The result of running the Principal Components analysis (in the Multivar menu).
Note that the Correlation radio button has been selected.
Software Availability
Use the PAST program. It is a free download (Windows PC only) and runs without the need for installation.
Example Problems (test your knowledge)
Note the ratios of the Eigenvalues (you have to calculate these yourself):
1:2 3.39042/2.94884 = 1.1497 (not significant as it is < 3)
2:3 2.94884/0.882822 = 3.340 (significant: look for subgroups)
Examine the structure of the subgroups by clicking on View loadings. Again, make sure that Correlations is chosen.
The apparent direction and magnitude of the bars in the Loadings plot shows two distinct groups: A-E and F-I. Also, it appears that the second group is more homogeneous. Don't expect that the groups will always be adjacent bars; that was just a characteristic of this particular set of data.
Page updated
Google Sites
Report abuse