Chapter 4 Summary

Chapter 4 Summary
Data and the Nature of Measurement

The focus of this chapter is on the process of observation and measurement in research.

Measurement

Measurement of a variable involves assigning numbers that represent values of the variable. The characteristics of the abstract number system do not always match the characteristics of the variable. The abstract number system has four characteristics: (1) identity, (2) magnitude, (3) equal intervals, and (4) a true zero. Identity means that each number has a unique meaning; magnitude means that the numbers have an inherent order; equal intervals means that the distance between two adjacent numbers is the same at all points on the scale; and a true zero means that the number zero represents none of the quality being measured.

Scales of Measurement

We use four scales of measurement that vary in how closely the scale matches the characteristics of the abstract number system.

Nominal scales have only the property of identity. They are used to classify participants into discrete categories, such as diagnostic group or political affiliation. Data from nominal scales are called nominal data (also sometimes called categorical data).
Ordinal scales have both the property of identity and magnitude. Participants can be ordered on their scores based on ordinal scales. Data from ordinal scales are called ordered data.
Interval scales have the properties of identity, magnitude, and equal intervals. Therefore, the difference between two scores is meaningful. But, since interval scales do not have a true zero, the ratio of two scores will not be meaningful. Data from interval scales are called score data.
Ratio scales have all of the properties of the interval scales as well as a true zero point. Therefore, all mathematical operations are appropriate, including taking the ratio of two numbers. For example, someone who earns $60,000 earns twice as much as someone who earns $30,000. Data from ratio scales are called score data.

Measuring and Manipulating Variables

Once variables are identified and defined in research, they must be measured and controlled. In order to do that, the researcher must understand concepts of measurement error and operational definitions.

Measurement Error

Measurement error can distort the scores so that the observations are no longer an accurate reflection of reality. Many factors can contribute to measurement error. Response set biases, which are the tendency for people to answer questions in biased and inaccurate ways, are one source of measurement error. One of the most powerful response set biases is social desirability (the tendency to respond to questions with the most socially acceptable answer). It is critically important to minimize measurement error because such error can attenuate (reduce) any observed relationship between two variables.

Operational Definitions

An operational definition is a definition of a variable in terms of the way the variable is to be measured or manipulated by the researcher. Carefully chosen operational definitions can help minimize many sources of measurement error.

Evaluating Measures

Developing measures by operationally defining variables is a critical first step, but researchers are also responsible for evaluating the quality of measure--that is, evaluating their reliability, effective range, and validity.

Reliability

Reliability refers to how reproducible our measure are--that is, do we get the same score each time we measure a given participant. There are different types of reliability. Interrater reliability measures the level of agreement between two independent observers. Test-retest reliability measures the stability of the variable over time. Internal consistency reliability measures the consistency or agreement of separate observations when the participant's score is the sum of those observations.

Effective Range

A factor to consider in selecting a measuring instrument or procedure is the effective range of the measure. Instruments can usually measure a variable accurately only within a certain range and will be relatively insensitive outside of that effective range. For example, a math test designed to evaluate second graders might be sensitive to differences in math ability in second graders, but may be too easy to discriminate the different levels of math skills in sixth graders.

In relation to the effective range of a measure, scale attenuation effects refer to restricting the range of a scale. Whenever the scores from participants tend to bunch near the top of the scale (ceiling effect) or the bottom of the scale (floor effect), the measure is incapable of effectively detecting differences among people. This can have serious consequences for research.

Validity

The validity of a measure refers to how accurately the measure reflects reality. Note that a measure must be reliable in order to be valid, but it is possible for a measure to show adequate reliability without being valid.

The Need for Objective Measurement

Objectivity is critical in science. Science depends on observations being reproducible. Subjective impressions of behavior are rarely reproducible and, therefore, it would be impossible to develop general behavioral principles on the basis of subjective impressions. Statistics provide a tool for the objective evaluation of a pattern of events. Consequently, objective measures and statistical analyses make it more likely that two researchers will agree on the significance of a given set of data.

Ethical Principles

The rare instances of deliberate distortion of data by scientists are serious attacks on the entire scientific enterprise. Although our society has always had a high tolerance for distorting facts (see advertisers and politicians, for example), such distortions are intolerable in science. They represent the most egregious ethical offense that a scientist can commit.