Metric and nonmetric data bases
mds index (research design issues)

Different nonmetric data obtained from the same set of stiumuli are likely to lead to different statistical results. This also holds true for different metric data obtained from the same sample population. For example, how many different metric and nonmetric standards can be employed to rank the city-pairs utilized in our examples on nonmetric data gathering? The number must be very large. An economist who thinks in terms of metric economic and financial data is surely to come up with similarity orderings different from those of a linguist using standards based on metric demographic and anthropological data. A psychologist or sociologist interested in gang formation or extended family relationships using either metric or nonmetric data is likely to come up with still another set of standards. Thus, it is not just a problem about the way in which data are gathered (metric or nonmetric), but also the sources used to obtain the data. Popular and scientific opinion often diverge, and many political decisions are based on what is popularly believed, rather than what is scientifically known. More important for the problem at hand, scientists often limit their research to data which is already available, rather than finding new data that better reflects the reality against which their hypotheses could be more appropriately tested.

top

Stimulus Set
mds index (research design issues)

As human perception is a complex phenomenon the researcher must select sets of stimuli whose elements are sufficiently different so as to evoke discrimination, but not so different that they open the entire panoply of human invention.

top

Pilot studies
mds index (research design issues)

As popular opinion is often elusive and the gathering of data expensive, it is important that researchers have a good idea about that for which they are looking before they begin. In the absence of previous studies such information may not be available, and the researcher sounds the population before he begins his search. Pilot studies allow researchers to test the power of their survey instruments before they test the entire population. An awful lot of time, effort, and money can be needlessly spent in the collection of data that is of little use to anyone, if a pilot study is not conducted in advance. Pilot studies can uncover poorly worded questions, inappropriate response alternatives, redundancy, and missing information. They can also reveal procedural weaknesses, identify potential sources of measurement error, and confirm or disconfirm popularly held beliefs that are crucial to the later gathering of information in the larger population.

top

Ideal elements
mds index (key terms | research design issues)

What is ideal to one respondent may not be ideal to another. This is especially problematic when the elements of the stimulus set are themselves controversial. As it is often the researcher's goal to understand the perceptual make-up of an entire population, rather than that of individual respondents, the notion of an ideal element can become very problematic. If the population is relatively homogeneous with regard to what the ideal element might be, a simple averaging of all respondents' responses will likely prove sufficient. If on the other hand, the population is unable to agree more than one ideal element is likely to arise. In this latter instance a simple averaging of ideal elements is likely to result in a false ideal. See Problems of segmentation below for further discussion.

top

Problems of segmentation
mds index (research design issues)

As it is not unusual to find important subdivisions in a large population, the likelihood of finding more than one ideal element is high. For example, youth and elderly are likely to view the same stimulus set from very different points of view. Managers and workers may or may not use the same standards of comparison depending on the nature of the stimulus set. Needless to say the number of possible subgroupings is enormous. Accordingly, the overzealous researcher, who simply lumps all respondents into the same perceptual bucket, is likely to come up with a perceptual framework that poorly reflects the decision-making framework of the population under study. In order to avoid this dilemma clustering techniques can be employed to identify important sub-populations before the multidiscriminant procedure is applied.

top

Learning effects
mds index (research design issues)

As the number of stimulus-pairs to be compared and/or ranked increases dramatically with each new stimuli added to the stimulus set, even small sets of stimuli can easily result in a very large number of comparisons. More comparisons result in more time spent and greater opportunity to learn on the part of respondents. In order to control for the likelihood of learning bias, as the respondent walks through the various combinations of stimulus pairs, a small number of pairs are repeated and later checked for learning effects. As learning effects can distort the data and thus the perceptual maps that result, it is important to control for these effects.

statsitical modelling (mds)

top

statistical modelling (diagnostics)