Chapter 12 Summary

Explain what constructs are, how they are developed, and why they are important to measurement and scale designs.

Within the overall process of creating meaningful information for resolving today's and future business/marketing problems, researchers must be able to develop appropriate questions and record the raw responses to those questions. Next to correctly defining the information problem, determining what type of data should be collected is the second most critical aspect in information research. Gaining access to raw data responses is achieved by the scale measurement incorporated into the questioning process. A construct can be viewed as any object that cannot be directly observed and measured by physical devices. Within the development process, researchers must consider the abstractness of the construct, its dimensionality, assessments of validity, and its operationalization. Not knowing exactly what it is that one needs to measure makes it difficult to design the appropriate scale measurements.

Discuss the integrated validity and reliability concerns underlying construct development and scale measurement.

Regardless of the method used for data collection researchers must strive to collect the most accurate data and information possible. Data accuracy depends heavily on the validity of the constructs and the reliability of the measurements applied to those constructs. Constructs can be assessed for content, convergent, discriminant, and nomological validity. Testing for reliability of constructs is indirectly achieved by testing the reliability of the scale measurements used in data collection. Scale reliability test methods available to researchers include test-retest, equivalent form, and internal consistency. Although scale measurements may prove to be reliable, reliability alone does not guarantee construct validity.

Explain what scale measurement it, and describe how to correctly apply it in collecting raw data from respondents.

Scale measurement is the process of assigning a set of descriptors to represent the range of possible responses that a person gives in answering a question about a particular object, construct, or factor. This process aids in determining the amount of raw data that can be obtained from asking questions, and therefore indirectly impacts the amount of primary information that can be derived from the data. Central to the amount of data issue is understanding that there are four basic scaling properties (i.e., assignment, order, distance, and origin) that can be activated through scale measurements. The rule-of-thumb is that as a researcher simultaneously activates more properties within the question/answering process, the greater the amount of raw data that can be classified into one of four mutually exclusive types: state-of-being, state-of-mind, state-of-behavior, and state-of-intention. Understanding the categorical types of data that can be produced by individuals' responses to questions improves the researcher's ability in determining not only what questions should be asked, but also how to ask those questions.

Identify and explain the four basic levels of scales, and discuss the amount of information they can provide a researcher or decision maker.

The four basic levels of scales are nominal, ordinal, interval, and ratio. Nominal scales are the most basic and provide the least amount of data. They activate only the "assignment" scaling property: the raw data do not exhibit relative magnitudes between the categorical subsets of responses. The main data structures (or patterns) that can be derived from nominal raw data are in the form of modes and frequency distributions. Nominal scales would ask respondents about their religious affiliation, gender, type of dwelling, occupation, or last brand of cereal purchased, and so on. The questions require yes/no, like/dislike, or agree/disagree responses.
Ordinal scales require respondents to express their feelings of relative magnitude about the given topic. Ordinal scales activate both the assignment and order scaling properties and allow researchers to create a hierarchical pattern among the possible raw data responses (or scale points) that determine "greater than/less than" relationships. Data structures that can be derived from ordinal scale measurements are in the forms of medians and ranges as well as modes and frequency distributions. An example of a set of ordinal scale descriptors would be "complete knowledge," "good knowledge," "basic knowledge," "little knowledge," and "no knowledge." While the ordinal scale measurement is an excellent design for capturing the relative magnitudes in respondents' raw responses, it cannot capture absolute magnitudes.

An interval scale activates not only the assignment and order scaling properties but also the distance property. This scale measurement allows the researcher to build into the scale elements that demonstrate the existence of absolute differences between each scale point. Normally, the raw scale descriptors will represent a distinct set of numerical ranges as the possible responses to a given question (e.g., "less than a mile," "1 to 5 miles," "6 to 10 miles," "11 to 20 miles," "over 20 miles"). With interval scaling designs, the distance between each scale point or response does not have to be equal. Disproportional scale descriptors (e.g., different-sized numerical ranges) can be used. With interval raw data, researchers can develop a number of more meaningful data structures that are based on means and standard deviations, or create data structures based on mode, median, frequency distribution, and range.

Ratio scales are the only scale measurements that simultaneously activate all four scaling properties (e.g., assignment, order, distance, and origin). Considered the most sophisticated scale design, they allow researchers to identify absolute differences between each scale point and to make absolute comparisons between the respondents' raw responses. Normally, though, the respondent is requested to choose a specific singular numerical value. The data structures that can be derived from ration scale measurements are basically the same as those for interval scale measurements. It is important to remember that the more scaling properties simultaneously activated, the greater the opportunity to derive more detailed and sophisticated data structures and therefore more information. Interval and ration scale designs are most appropriate to use when researchers want to collect either state-of-behavior, or state-of-intentions, or certain types of state-of-being data.

Discuss the ordinally-interval hybrid scale design and the types of information it can provide researchers.

Some researchers misidentify certain types of ordinal scales as being interval scaled. They take an ordinal scale design and artificially assume that the scale has activated the distance (and the origin) scaling properties. This assumption comes about when the researcher arbitrarily assigns a secondary set of numerical scale descriptors (e.g., consecutive whole integers) to the original primary set of ordinal descriptors. What drives a researcher to misrepresent the ordinal scale format is the need to know not only an individual's attitudes or feelings but also the combined overall attitude or feeling of a group of individuals. To achieve this overall response outcome, the researcher must be able to add together many separate raw responses and perform some type of basic mathematical procedure, like establishing the mean response of the group. There are two main approaches to developing an ordinally-interval scale measurement: (1) using a secondary set of cardinal number descriptors and redefining the complete set of primary scale descriptors (1= definitely agree, 2= generally agree, 3= slightly agree, 4=slightly disagree, 5=generally disagree, and 6= definitely disagree); or (2) using primary descriptors to identify only the extreme end points of a set of secondary cardinal numbers that make up the range of raw scale descriptors, or scale points (definitely agree 1 2 3 4 5 6 definitely disagree).

Regardless of the method used, for the researcher to believe that the absolute difference between a respondent response of "definitely agree" and that of another respondent's "generally agree" is one unit of agreement is really nothing more than a leap of faith. Researchers are cautioned to be very careful how they interpret the data structures generated from this hybrid scale design.

Discuss three components of the scale development and explain why they are critical to gathering primary data.

In developing high-quality scale measurements, the researcher must understand that there are three critical components to any complete scale measurement; question/setup; dimensions of the object, construct, or behavior; and the scale point descriptors. Some of the criteria for scale development are the intelligibility of the questions, the appropriateness of the primary descriptors, the discriminatory power of the scale descriptors, the reliability of the scale, the balancing of positive/negative scale descriptors, the inclusion of a neutral response choice, and desired measures of central tendency (mode, median, and mean) and dispersion (frequency distribution, range, estimated standard deviation). If the highest-quality raw data is to be collected to transform into useful primary information, researchers and practitioners alike must have an integrated understanding of construct development and scale measurement.