What Is Measurement?

1 1 What Is Measurement? OverviewThis chapter covers the basics of measurement . The chapter aims to provideunderstanding of measurement and measure development procedures cur-rently discussed in the literature. measurement error is introduced, and thesteps in the measure development process and empirical procedures to assesserror are described. The chapter uses many examples based on data to illus-trate measurement issues and procedures. Treatment of the material is at anintuitive, nuts-and-bolts level with numerous illustrations. The chapter is not intended to be a statistics primer; rather, it providessufficient bases from which to seek out more advanced treatment of empiri-cal procedures. It should be noted that many issues are involved in usingappropriately the statistical techniques discussed here. The discussion in thischapter aims to provide an introduction to specific statistical techniques, andthe references cited here provide direction for in-depth Is measurement Error?

2 measurement is essential to empirical research. Typically, a method used tocollect data involves measuring many things. Understanding how any onething is measured is central to understanding the entire research measurement has been defined as rules for assigning numbers toobjects in such a way as to represent quantities of attributes ( , Nunnally,1978, p. 3). measurement consists of rules for assigning symbols to 1/10/2005 2:51 PM Page 1so as to (1) represent quantities of attributes numerically (scaling) or(2) define whether the objects fall in the same or different categories withrespect to a given attribute (classification) (Nunnally & Bernstein, 1994,p. 3).1 The attributes of objects, as well as people and events, are the under-lying concepts that need to be measured. This element of the definition ofmeasurement highlights the importance of finding the most appropriateattributes to study in a research area.

3 This element also emphasizes under-standing what these attributes really mean, that is, fully understanding theunderlying concepts being measured. Rules refer to everything that needs tobe done to measure something, whether measuring brain activity, attitudetoward an object, organizational emphasis on research and development, orstock market performance . Therefore, these rules include a range of thingsthat occur during the data collection process, such as how questions areworded and how a measure is administered. Numbers are central to the def-inition of measurement for several reasons: (a) Numbers are standardized andallow communication in science, (b) numbers can be subjected to statisticalanalyses, and (c) numbers are precise. But underneath the fa ade of precise,analyzable, standardized numbers is the issue of accuracy and very idea of scientific measurement presumes that there is a thingbeing measured ( , an underlying concept).

4 A concept and its measurementcan be distinguished. A measure of a concept is not the concept itself, butone of several possible error-filled ways of measuring distinction canbe drawn between conceptual and operational definitions of conceptual definition describes a concept in terms of other concepts(Kerlinger, 1986; Nunnally, 1978). For instance, stock market performanceis an abstract notion in people s minds. It can be defined conceptually interms of growth in value of stocks; that is, by using other concepts such asvalue and growth. An operational definition describes the operations thatneed to be performed to measure a concept (Kerlinger, 1986; Nunnally,1978). An operational definition is akin to rules in the definition of mea-surement discussed earlier in the chapter , and refers to everything that needsto be done to measure something. The Dow Jones average is one measureof stock market performance .

5 This operational definition involves trackingthe stock value of a specific set of companies. It is by no means a perfectmeasure of stock market performance . It is one error-filled way of measur-ing the concept of stock market term constructis used to refer to a concept that is specifically definedfor scientific study (Kerlinger, 1986). In Webster s New World Dictionary,constructmeans to build, form or devise. This physical meaning of theword constructis similar to the scientific meaning: Constructs are concepts2 measurement Error and Research 1/10/2005 2:51 PM Page 2devised or built to meet scientific specifications. These specifications includeprecisely defining the construct, elaborating on what it means, and relatingit to existing research. Words that are acceptable for daily conversationwould not fit the specifications for science in terms of clear and precisedefinitions. I am going to study what people think of catastrophic events is a descriptive statement that may be acceptable at the preliminary stagesof research.

6 But several concepts in this statement need precise description,such as think of, which may separate into several constructs, and catas-trophe, which has to be distinguished from other descriptors of explanations are essentially words, and some of these words relateto concepts. Constructs are words devised for scientific science,though, these words need to be used carefully and defined measuring something, error is any deviation from the true value,whether it is the true value of the amount of cola consumed in a period of time,the level of extroversion, or the degree of job this truevalue is rarely known, particularly when measuring psychological variables,this hypothetical notion is useful to understand measurement error inherent inscientific research. Such error can have a pattern to it or be all over theplace. Thus, an important distinction can be drawn between consistent ( ,systematic) error and inconsistent ( , random) error (Appendix ).

7 Thisdistinction highlights two priorities in minimizing error. One priority is toachieve consistency,5and the second is to achieve explanations of random and systematic error are provided below,although subsequent discussions will introduce nuances. Consider using aweighing machine in a scenario where a person s weight is measured twicein the space of a few minutes with no apparent change (no eating and nochange in clothing). If the weighing machine shows different readings, thereis random errorin measurement . In other words, the error has no pattern toit and is inconsistent. Alternatively, the weighing machine may be off in onedirection, say, consistently showing a reading that is 5 pounds higher thanthe accurate value. In other words, the machine is consistent across readingswith no apparent change in the weight being measured. Such error is calledsystematic errorbecause there is a consistent pattern to it. It should be noted,though, that on just one reading, the nature of error is not clear.

8 Even if thetrue value is independently known through some other method, consistencystill cannot be assessed in one reading. Multiple readings suggest the incon-sistent or consistent nature of any error in the weighing machine, providedthe weight of the target person has not changed. Similarly, repetition eitheracross time or across responses to similar items clarifies the consistent orinconsistent nature of error in empirical measurement , assuming the phenome-non across time is Is measurement ? 1/10/2005 2:51 PM Page 3If the weighing machine is all over the place in terms of error ( ,random),conclusions cannot be drawn about the construct being error in measures attenuates relationships (Nunnally, 1978); thatis, it restricts the ability of a measure to be related to other measures. Thephrase all over the place is, in itself, in need of scientific precision, whichwill be provided in subsequent pages. Random error has to be reducedbefore proceeding with any further analyses.

9 This is not to suggest norandom error at all, but just that such error should be reasonably , a viable approach is to collect multiple observations of such read-ings in the hope that the random errors average out. Such an assumptionmay work when random errors are small in magnitude and when the mea-sure actually captures the construct in question. The danger here, though, isthat the measure may not capture any aspect of the intended , the average of a set of inaccurate items remains of abstract concepts in the social sciences, such as attitudes, arenot as clear-cut as weighing machines. Thus, it may not be clear if, indeed,a construct, such as an attitude, is being captured with some random errorthat averages out. The notion that errors average out is one reason for usingmultiple items, as discussed that are relatively free of random error are called reliable(Nunnally, 1978). There are some similarities between the use of reliabilityin measurement and the common use of the term.

10 For example, a personwho is reliable in sticking to a schedule is probably consistently on time. Areliable friend is dependable and predictable, can be counted on, and is con-sistent. However, there are some differences as well. A reliable person whois consistent but always 15 minutes late would still be reliable in a measure-ment context. Stated in extreme terms, reliability in measurement actuallycould have nothing to do with accuracy in measurement , because reliabilityrelates only to consistency. Without some degree of consistency, the issue ofaccuracy may not be germane. If a weighing machine is all over the place,there is not much to be said about its accuracy. Clearly, there are shades ofgray here in that some small amount of inconsistency is acceptable. Waitingfor perfect consistency before attempting accuracy may not be as efficient orpragmatic as achieving reasonable consistency and approximate consistency may not even be possible in the social sciences, given theinherent nature of phenomena being a measure is accurate or not is the realm of validity,or thedegree to which a measure is free of random and systematic error (Nunnally,1978).