Multidimensional Scaling

DISCLAIMER: All of my programs are free of charge. That being said, the user assumes all liability for the proper use and any complications pertaining to these files. I hope you can put them to good use!

 

SpAM

Spatial Arrangement Method (SpAM) of Multidimensional Scaling: This is an experiment designed to emulate the spatial arrangement method of multidimensional scaling (originally proposed by Rob Goldstone, 1994, Behavior Research Methods Instruments and Computers).  Multiple stimuli are presented simultaneously, and the participant moves the stimuli around the screen (using drag-and-drop), placing them at distances that are proportional to the user’s subjective similarity ratings (please see the Publications tab of this website for more information on the use of this method).  The output file contains a set of euclidean distances of each item to every other (large distances indicate high dissimilarity).

The ZIP file that you download contains two experiments: one for use with images, and another for use with text.  All image files are included in the download.  Also, there are sample data files and screen-grabs (basically, pictures created automatically of individual subject’s solutions) included in the ZIP.  Please see below for detailed instructions on how to use and modify the experiment.  See also the MDS Macros sub-section of the Software page of this site, for macros that will transform your data into matrices that can be directed copied into statistical software packages like SPSS.  If you utilize this software for published research, we would appreciate if you cited our article.  Thank you!  Download the files here: E-Prime_SpAM.

Here is the citation:

Hout, M. C., Goldinger, S. D., & Ferguson, R. W. (2013). The versatility of SpAM: A fast, efficient spatial method of data collection for multidimensional scaling. Journal of Experimental Psychology: General, 142, 256-281.

Spatial Arrangement Method of Multidimensional Scaling, Instructions:  This document will tell you everything you need to know in order to successfully adapt the Spatial Arrangement experiment to your needs.  Download the instructions here: SpAM_Instructions.

 

SpAM_JAVA

JAVA application for the Spatial Arrangement Method (SpAM): This is a version of the spatial arrangement method provided for those who do not use E-Prime. It has all the same functionality as the basic SpAM technique, allowing the user to collect single-trial SpAM experiments with images or text. Please refer to the instructions provided below. Also, I am also happy to provide the source code for this experiment, in case the user wants to modify the experiment in any way. If you make any useful modifications to the code, please feel free to share it with me and I will host a library of JAVA applications for this method. Please also make sure to cite us if you make use of this method for any published work (see above).  We would like to thank Sathananda Siloju for his help in programming this application!

The ZIP file that you download contains several files. First, the Hout_SpAM_JAVA.JAR file is the executable JAVA application. This will run the experiment for you. Unlike the E-Prime version, you do not have to specify starting locations for your items. The program will automatically generate starting positions for you. The folder contains some sample image files so you can see how the program runs with pictures, and it also contains a sample text file with word stimuli, so you can see how it is implemented with text. Finally, there is a sample data file so you can familiarize yourself with the output. As with the E-Prime version, the data is simply 3 columns of information. Two containing the names of each image being compared, and one with the final Euclidean distance between the items (measured in pixels). Please refer to the descriptions above for a more thorough treatment of the SpAM technique.  See also the MDS Macros sub-section of the Software page of this site for Excel VBA macros that will help organize your data for analysis in software applications such as SPSS.  Download the files here: SpAM_JAVA.

JAVA version of SpAM, Instructions: This document will tell you everything you need to know in order to execute the JAVA application.  Download the instructions here: SpAM_JAVA_Instructions.

JAVA version of SpAM, Source Code: This is the source code for the JAVA application. All files are included, so the user can modify the application to fit their needs.  Download the files here: SpAM_JAVA_Source.

 

SpAM_MultiTrial

SpAM method of Multidimensional Scaling, multiple-trial technique: This experiment is an advanced version of SpAM that can be used when you want to scale more items than you can fit on the screen at one time.  Perhaps you want to scale more than 30 items, or you have images that are very large (and so cannot fit many onscreen at once).  No matter what the reason, this method will show you a technique you can use to scale larger sets of stimuli across several trials.  Ideally, you want to pair each item with every other once, and only once. However, depending on the parameters of your experiment (i.e., how many total stimuli you have, how many items you display at once), that may or may not be possible.

The general method I suggest is to employ a Steiner System. A Steiner System is a mathematical tool that you can use to control which items are shown on which trials. Depending on the parameters of your experiment (i.e., the combination of total stimuli, and items shown per trial), there may be a Steiner System that does not repeat pairwise comparisons. That is, a set of trials that will show each item in comparison with every other only once. For other combinations, that may not be possible, and some stimuli may have to be shown with others more than once. It’s an imperfect system, leading to (per subject) multiple observations per “cell” (i.e., per pairwise comparison). However, as you’ll see with the program and instructions I have provided, we can roughly account for this issue by assuring that, across subjects, the particular pairings that are repeated get randomized. With enough subjects, that would give the user an overall matrix with equal (or approximately equal) observations per cell.

In the sample I provide, I show you how to scale 35 images in one experiment, sampling 15 items per trial. This takes a total of 7 trials to complete, and some pairings are shown more than once. The vast majority of pairings, however, are only shown once. The ZIP file that you download contains a sample experiment, a sample data file, and the resources necessary to try this out for yourself. There is also an Excel workbook that will allow you to concatenate your data into a matrix that can be copied directly into statistical software packages. Again, if you use this software, please cite our JEP:General paper listed above.

Finally, one alternative to using the Steiner System design would be to simply allow random sampling to populate individual trials. For instance, suppose you have 20 stimuli, and you want to show 5
items per trial. With 4 total trials, each subject would have seen each image exactly once. Of course, that does not mean that each cell of your design (i.e., each pairwise comparison) has been filled.
If Item #1 is shown on the first trial, and Item #4 is shown on the second trial, then you will not have a direct comparison of those two items. Across subjects, however, the cells of your design may
fill in, provided you have a large enough data set. In the MDS Macros sub-section, I have provided a worksheet that allows you to simulate how many subjects you would need to fill a randomly populated design with whatever parameters you so choose (e.g., total number of stimuli, number of items per trial, desired number of observations per cell, etc.).  Download the files here: SpAM_MultiTrial.

SpAM Method of Multidimensional Scaling, Multiple-Trials Instructions: This document will tell you everything you need to know in order to successfully adapt the multi-trial version of the Spatial Arrangement experiment to your needs.  Download the instructions here: SpAM_MultiTrial_Instructions.

 

 Pairwise

Classic (Pairwise) Multidimensional Scaling: This experiment runs the classic, pairwise comparison method of multidimensional scaling. The subject is presented with stimuli (images or text) 2 at a time, and makes a subjective similarity decision on a scale from 1-9. The user simply indicates how many stimuli are to be scaled, and whether the stimuli are images or text (see below for detailed instructions on how to use the experiment). The code will automatically present subjects with every possible pairwise comparison in randomized order (with random designation of stimuli to the left or right side of the screen). The output file contains the subject’s rating for every comparison (see the MDS Macros sub-section of the site for code that will automatically create individual subject’s MDS matrices from this data, which can be copied directly into statistical software packages). The ZIP file that you download contains the experiment and resources necessary for demonstration purposes.  Download the files here: Pairwise_MDS.

Classic (Pairwise) Multidimensional Scaling, Instructions: This document will tell you everything you need to know in order to successfully adapt the Classic MDS experiment to your needs. Download the instructions here: PairwiseMDS_Instructions.

 

TotalSet

 Total-Set (Pairwise) Multidimensional Scaling: This experiment involves a method very similar to the classic, pairwise comparison method of multidimensional scaling. However, rather than present the subject with only two items at a time, the entire stimulus set is shown, and two items are highlighted. Like traditional pairwise methods, the subject judges the stimuli (images or text) subjectively, on a scale from 1-9. But unlike traditional pairwise methods, wherein the first several ratings may be arbitrary, this method places each comparison in the context of the entire stimulus set from the very outset. To utilize this method, the researcher need make few changes to the code. You will indicate how many stimuli are to be scaled, and whether you intend to use images or text (see below for detailed instructions on how to use the files). The code will automatically present subjects with every possible pairwise comparison in randomized order. The output file contains the subject’s rating for each comparison (see the MDS Macros sub-section of the site for code that will automatically create individual subject’s MDS matrices from this data, which can be copied directly into statistical software packages). The ZIP file that you download contains the experiment and resources necessary for demonstration purposes.  Download the files here: TotalSet_MDS.

Total-Set (Pairwise) Multidimensional Scaling, Instructions:vThis document will tell you everything you need to know in order to successfully adapt the Total-Set MDS experiment to your needs.  Download the instructions here: TotalSetMDS_Instructions.