Application of PINDIS separately in Hamlet II

By Avishek Majumder & Priya Chetty on October 25, 2018

The previous article showed steps to perform Procrustean Individual Differences Scaling (PINDIS) using ‘Select’ method. This article explains application of PINDIS separately in Hamlet II. It also presents an example using PINDIS analysis to understand the application in depth. Accessing PINDIS separately is possible only after creating the input file using ‘Select’ function.

Steps to access PINDIS separately

Follow the steps provided in the last article to save the input file (“.inp”) in order to access the PINDIS separately.

Thereafter follow the below steps in Hamlet II.

  • In the toolbar, go to ‘MDS’.
  • Click on ‘PINDIS- Procrustean Individual Differences Scaling’.
Figure 1: Selection option for PINDIS in Hamlet II
Figure 1: Selection option for PINDIS separately in Hamlet II

Furthermore, follow the below four steps.

  1. Select the number of dimensions to be shown in the resulting graph (step 1 in the figure above).
  2. Click on ‘…’ to browse and select the saved file (step 2 in the figure above).
  3. Click on ‘Open’ (step 3 of the figure above).
  4. If a prompt by a pop-up ‘By default, the centroid will be used for reference’ shows up, then click on ‘Yes’.

Another dialogue box will appear which will confirm the configuration. Click on ‘Fit data to this configuration’ to continue to the analysis. Some of the users might receive a warning box stating ‘The viewpoint is not at the origin! ZOOM out to continue’. Keep clicking on ‘OK’ till it zooms out automatically in the background and optimizes the shape accordingly.

How to edit the graph?

In a new output window, a three-dimensional graph will appear with all the text points plotted based on their relationship according to the centroids (figure below). Use different functions in the task bar below in order to edit the graph.

Figure 2: How to edit the graph
Figure 2: How to edit the graph

Performing PINDIS analysis

In this section, PINDIS analysis is performed using an example to illustrate the use and significance of this type of multi-dimensional scaling analysis. For this example, use the following Michigan-Nijmegen Integrated Smallest Space Analysis (MINISSA in Hamlet II for multidimensional scaling) output file.

Dimension 1 2 3
1 context* -0.245 0.829 -0.415
2 dimension* 1.385 0.497 -0.105
3 frequenc* -0.003 -0.383 -0.627
4 hamlet -0.115 0.28 0.734
5 joint 0.048 -0.137 -0.553
6 MINISSA 0.826 -0.655 0.264
7 scaling 0.851 -0.339 0.25
8 text* -0.555 0.362 0.401
9 vocabulary -1.39 -0.371 0.026
10 word* -0.803 -0.084 0.025

The hypothesis configurations will emerge for this data. In the figure below, these derived results have been rescaled based on the biggest coordinate value available to make it easier to view and understand. This also includes changes in sample text file to match the parameters with the contrasting text file for simpler analysis.

Figure 3: Plot for coordinate values from PINDIS separately in Hamlet II
Figure 3: Plot for coordinate values from PINDIS separately in Hamlet II

Findings of subject space graphical representation

Subject space shows the relationship of the different transcripts under comparison and provides normalized dimension weights of the similarity matrices. It is in furtherance to the results of MINISSA sets. The figure below exhibits the subject space where the points are present in a way that they are nearest to the centroid.

Click on ‘Data’ and then ‘View or edit data’ in order to view the normalized dimension weights of the graph.

Figure 4: Subject space graphical representation
Figure 4: Subject space graphical representation
Normalized Dimension Weights
Subject Communality 1 2 3
1 **** hamlet 0.9891 0.7902 0.4428 0.4106
2 **** babel 0.1951 0.0686 0.1271 0.4174

The above ‘Normalized Dimension Weights’ show the placement and plotting of each of these subjects on different dimensions in Subject Space as shown in the figure above.

Click on ‘Next’ in order to view the results for different PINDIS transformations of each subject individually. Right-click on the endpoints of the labels to view a new graph on individual information.

Figure 5: Three dimensional plotting of the similarity transformations
Figure 5: Three-dimensional plotting of the similarity transformations

Generate the two individual configurations in the image above by clicking on ‘Next’. These configuration graphs will display the three-dimensional plotting of the similarity transformations of each of the texts available in individual files. The table below emerges after testing the similarity of each of the texts based on their relative centroids. Each of the dimensions is showcasing closely related texts by assigning significant weights accordingly.

Similarity Transformations
Normed scalar for unconditional weights:  1.000000
1 context* -0.0737 0.2686 -0.1199
2 dimension* 0.4403 0.152 -0.0273
3 frequenc* -0.003 -0.1126 -0.2032
4 hamlet -0.0347 0.0792 0.2357
5 joint 0.0144 -0.0361 -0.1766
6 MINISSA 0.2582 -0.2143 0.0742
7 scaling 0.2676 -0.1144 0.074
8 text* -0.1736 0.1117 0.1319
9 vocabulary -0.4413 -0.111 0.0041
10 word* -0.2543 -0.0231 0.0072
Fit of subject to hypothesis <P0> — S (Z, X) = 0.820667

Optimal multidimensional scaling results using PINDIS

There are various multidimensional scaling techniques available but PINDIS is one of the highly sophisticated procedures that applies a series of variations. This is because it increases flexibility in each iteration in order to obtain optimal results. Although this the last test of the Hamlet II software, there are more tools or software for similar analyses. Therefore the next article will review other text-based quantitative analysis software.

Discuss