Rmppp ch04web

4
Measurement theory
and practice
Rajender R Aparasu
Chapter objectives
* To explain the concept of measurement
* To discuss levels of measurement
* To describe reliability and methods to evaluate reliability
* To describe validity and methods to evaluate validity
Introduction
Measurement forms the basis for empirical health services research including
pharmaceutical practice and policy research. The measurement process is
designed to record and capture the underlying construct and concept. In
healthcare, critical policy and practice discussions are based on cost, quality,
and access. These concepts or constructs have to be measured and analyzed
using variables to evaluate the performance of a healthcare system at the
patient, family, institution, and population levels. Each of these constructs
is complex and often requires multiple measures or variables to capture the
underlying concepts. Advances in measurement of each of these dimensions
have led to improvement in the healthcare system. For example, the develop-
ment of the Diagnostic and Statistical Manual of Mental Disorders (DSM),
diagnostic-related groups (DRGs), and Healthcare Common Procedure
Coding System (HCPCS) has led to changes in healthcare delivery and reim-
bursement. In recent years, measurement of quality of healthcare based on
structure, process, and outcome has gained significance due to its importance
in healthcare delivery (Donabedian 2003).
The measurement process involves a systematic assignment of values or
numbers to observations based on a priori rules of measurement
(Viswanathan 2005). It is a critical step in quantitative research because it
also defines the subsequent steps in conducting research such as analysis and
Sample chapter from Research Methods for Pharmaceutical Practice and Policy

interpretation of the research findings. The measurement process involves
recording observations that are manifestations of the underlying construct;
this requires a good understanding of the construct and the measurement
process. Often the variables are operationalized based on the methodology
used to capture these variables. This can involve collection of data by the
researcher, also called primary data, or data collected by others for reuse,
also referred to as secondary data. The operationalization and collection of
data are critical in the measurement process. Further evaluation of this
measurement process will confirm whether the process is truly measuring
the construct, which entails ensuring that the measurement process is reliable
and valid. This chapter describes the concept and levels of measurement, and
discusses reliability and validity, and methods commonly used to evaluate
reliability and validity.
Nature and level of measurement
The measurement process for any construct is based on the existing knowl-
edge base regarding the construct. The measurement of physiological con-
structs such as blood pressure and blood glucose is often standardized.
However, behavioral constructs such as compliance are rather complicated
because the observed behavior may or may not reflect the intended construct.
An understanding of the measurement process for behavioral constructs can
provide a good framework to understand the measurement process in
general. Summers (1970) suggests that the measurement process of an
abstract construct such as behavior that is not directly observable involves
three interlinked operational subprocesses:
1 Identification of acceptable behavior specimens that represent the
underlying construct
2 Data collection of specimens
3 Conversion of specimens to a quantitative variable.
An acceptable specimen defines the data collection process, which in turn
determines the type of quantitative variable. Often operational definitions,
especially measured operational definitions, are based on these three opera-
tional subprocesses.
Acceptable behavioral specimen
Identification of acceptable specimens that represent the underlying construct
is a critical step in the measurement process. The current knowledge base
defines ones that are acceptable and those that are not. Acceptable specimens
for diagnosis of diabetes such as blood glucose and glycated hemoglobin are
often standardized. The measurement process for constructs such as
56 | Research Methods for Pharmaceutical Practice and Policy

compliance is complex. However, the concept of compliance with a medica-
tion regimen is important in evaluating treatment effects. Compliance refers
to the extent to which the patient follows healthcare advice (Hayes et al.
1979). Not all patient behaviors that reflect compliance with a medication
regimen are easily observable by researchers and clinicians. Compliance with
a medication regimen can include behaviors that can capture the extent to
which the patient is taking medication with respect to frequency and duration
of therapy. The behavior specimens such as refill history which capture fre-
quency and duration of therapy are considered acceptable for measuring
compliance (Farmer 1999). Other medication-taking behaviors can include
time of administration and avoidance of certain foods.
Data collection
The data collection process entails capturing data based on specimens. Often
multiple data collection processes can be linked to the behavior specimens.
With respect to compliance, the data collection can involve self-reports of
specimen behavior or observation of overt behavior such as refill history
(Farmer 1999). The self-report methodology involves measurement of patient
responses to a series of questions or items using a survey instrument to
determine medication-taking behaviors. Self-report methodology is based
on the assumption that the patients can report their medication use behavior.
Overt behaviors of compliance can be assessed by direct observation of
patient behavior or refill history as captured in prescription claims data.
Refill history measures, such as medication possession ratio, reflect the num-
ber of doses filled by the patient for the dispensing period. Refill history is
based on the assumption that prescription-filling behavior reflects the
patient’s medication use pattern. Compliance measurement based on bio-
chemical instruments to evaluate drug levels can also be used as a behavior
specimen (Farmer 1999).
Assignment of values
Converting the behavioral specimen to a variable involves assignment of
values based on the rules developed for data collection to evaluate the under-
lying concept. The assignment process should reflect variation in the under-
lying construct. For example, the value assigned to a compliance measure,
based on self-reports or refill history, should reflect the extent to which the
patient is following the healthcare advice for medications. Measurement and
further analysis are dependent on the properties of the variable. The levels of
measurement can be classified as nominal, ordinal, interval, and ratio, based
on the properties of the quantitative variable. This is also referred to as
Stevens’s (1946) scales of measurement.
Measurement theory and practice | 57

The nominal measure, also called the categorical measure, is the simplest
and lowest form of measurement. It is used for naming or identification
purposes. Examples include measurement of gender (male or female) and
ethnicity (white, black, etc.). The subgroups in a nominal variable are mutu-
ally exclusive and exhaustive. All members of a group have the same identity
and there is no relationship among the subgroups. For nominal measures the
assignment of numbers, letters, or symbols is only for labeling and grouping
purposes. Although numbers are often used for labeling or coding purposes,
they do not have an arithmetic value. Consequently, none of the mathematical
manipulations can be used for nominal measures. The nominal measures are
often used for counting and to examine frequency distribution. The number of
subgroups for classification or labeling is based on the extent of identification
needed. For example, the respondents’ residences can be grouped by state (50
states) or by region (northeast, midwest, south, and west). Nominal variables
can be used as dichotomous variables where only two subgroups are recorded
such as white and nonwhite for race.
The ordinal measure is rank ordering of a group membership with prop-
erties of transitivity. The group membership can be ordered from lowest to
highest with a clear interrelationship of the levels, unlike the nominal mea-
sure. For example, the health status of a patient can be measured with levels or
responses of excellent, very good, good, fair, or poor. Similarly, a patient’s
perception of compliance can be measured using a five-point scale from most
compliant to least compliant. Ordinal measures are often used in survey
instruments, and can be ordered from highest (excellent) to lowest (poor).
In addition, the rank order relationship levels are mutually exclusive and
exhaustive, similar to the nominal measure. In addition the interrelationship
of the levels is known. The distance between the levels is, however, not equal
and, at times, not known. The levels have properties of transitivity, i.e., if
excellent is better than very good and very good is better than good, this
makes excellent better than good.
The interval and ratio measurements have properties of equality of inter-
vals in addition to characteristics of an ordinal measure. The key difference
between interval and ratio measures is that there is an absolute zero level in
ratio measure. Ratio is the highest form of measurement because it can
represent the zero value of the measure. Mathematical manipulations such
as multiplication and division can be used for ratio measures. Examples of
ratio measures include prescription expenditure, hospitalization days, wait
time, and weight gain. The best example of an interval measure is temperature
measured on the Fahrenheit scale. The temperature is measured using the
equidistant scale of Fahrenheit in which there is no zero amount or quantity
of temperature; the assignment of zero temperature is arbitrary. The intelli-
gence quotient (IQ) scale is another example of an interval measure where the
distance between the scale values is the same. In psychometric research, it is

difficult to find a measure that has an absolute zero. The interval measures
allow mathematical manipulations. In general, ratio and interval mea-
sures are used in the same way for statistical analysis. Ratio and interval
measures provide great flexibility for mathematical manipulations as long
as the original properties of measurement are satisfied. With respect to com-
pliance, the number of missed doses is a ratio scale that can be obtained from
patients using self-reports.
Measurement issues
Ratio and interval measurements are the most desirable measures in health
services research because they capture more information about the underlying
construct, and are therefore more likely to reflect the variation in the under-
lying construct, than other ordinal or nominal measures. Ratio and interval
scales also have all the necessary properties required for mathematical manip-
ulations. Most importantly, these measures can be converted to low forms of
measures such as ordinal measures. It is not possible to convert ordinal
measures into ratio/interval measures. For example, family income captured
in dollars (ratio measure) can be grouped into high, middle, or low income
(ordinal measure) based on specific ranges of family income. However, it is
not possible to convert family income captured as an ordinal measure into a
ratio measure. In general, higher forms of measures should be preferred to
lower forms of measures.
The goal of the measurement is to assign values to a variable based on
the specific rules formulated to measure the underlying construct or con-
cept. The measurement process is also designed to capture variation in the
underlying construct. The values assigned in the measurement process are
distinct due to the differences in the underlying construct or concept. If
variations in the underlying construct or concept are not reflected in the
measurement, it leads to measurement error, which is an error in the mea-
surement process (Viswanathan 2005). Sources of measurement error can
be due to the measurement process itself or factors outside the measurement
process. For example, a measurement error may occur if the number of
missed doses is captured using self-reports in the presence or absence of a
physician. A good measurement process is designed to reduce such mea-
surement errors. These issues are considered in detail in the discussion of
reliability and validity.
Measurement and statistical analyses are based on the concept of analysis
of the variation. The type of measurement of variables determines the statis-
tical analysis to be used. This holds true for both dependent and independent
variables. The interval and ratio measures capture greater variation in the
dependent measures than the ordinal measures. Consequently, there is more
opportunity to explain the extent of variation in interval and ratio measures

than ordinal measures. This can also improve to capture the sources of var-
iation in statistical analysis.
Reliability and validity
The goal of the measurement process is to ensure that the values assigned to
variables are reliable and valid. Reliability and validity are different dimen-
sions of the measurement process. Reliability ensures that the assignment of
values is consistent or reproducible, whereas validity ensures that the assign-
ment of values truly reflects the underlying construct or concept (Bohrnstedt
1970; DeVellis 1991; Trochim 2001). Both reliability and validity are impor-
tant in the measurement process because reproducibility of a measure as well
as the trueness of a measure is critical in research. Reliability of a measure
does not, however, ensure its validity. For example, use of self-reports to
capture compliance with medication may be reliable but may not be valid.
Self-reports of missing doses may consistently provide the same measures but
may not truly capture the patient behavior. The constructs in sociobeha-
vioral research are often abstract and hence require evaluations of reliability
and validity of the measurement process. Psychometric research has played a
significant role in the evaluation of reliability and validity. Some of the
reliability and validity evaluation methods are specific to survey research
involving a survey instrument. However, the concepts of reliability and
validity are relevant for both behavioral and nonbehavioral constructs.
Measurement errors
Both reliability and validity of a measurement are affected by measure-
ment errors. Measurement errors can be classified as random errors and
nonrandom errors (Viswanathan 2005). Random or chance errors, as the
term suggests, occur inconsistently, and cause the measures to deviate
randomly from the true value. Random errors negatively affect the reli-
ability of the measurement and are present in every measurement process.
The goal of a measurement process is to minimize random errors and,
thereby, maximize the reliability. The factors that influence reliability
can be the individual, instrumentation, and environment. Individual or
patient level factors include diurnal variation, education level, or bio-
logical variability. Instrument level factors include calibration of instru-
ment, or misreading or mistakes in recording questionnaire responses.
Environmental factors influence individual and instrumental factors, such
as temperature, pressure, light, or electrical fluctuations. An ideal mea-
surement process will minimize the influence of these factors to maximize
the reliability. This can be achieved using standard and consistent data
collection and administration methods. For example, the measurement can

be recorded at one specific time for all participants to minimize diurnal
variation.
Nonrandom or systematic errors occur consistently by definition and
hence cause the measures to deviate from the true value nonrandomly
(Viswanathan 2005). The amount of systematic error directly influences the
validity of the measurement. It is inversely related to validity: high non-
random error decreases the validity of a measurement process. Systematic
bias is a classic example of a nonrandom error that threatens the validity of
a measurement process. Random bias does not influence the validity of a
measurement.
There are different types of biases such as information, recall, and inter-
viewer bias. Information bias occurs when there is consistently differential
information among the participants of interest due to underlying factors. For
example, in epidemiological research, test cases tend to provide more infor-
mation than control cases that do not have the disease. Recall bias occurs
where there is a differential ability to recall information about previous
experience due to issues related to time or experience. For example, the
measurement error in reporting the number of missed doses is likely to be
less than the number of doses actually taken according to the instructions
due to recall bias. Interviewer bias exists when interviewer perception or
behavior influences the responses. The measurement process should prefer-
ably control for systematic biases at both data collection and study design
stages to strengthen the validity. For example, blinding techniques are used
to hide the procedural aspects of study design from respondents and inter-
viewers to minimize the bias.
Reliability
Reliability, as discussed earlier, ensures that the measurement is consistent or
reproducible (Bohrnstedt 1970; DeVellis 1991; Trochim 2001). A reliable
measurement process will consistently provide the same or a similar value
for an unchanged variable, whereas a change in the underlying construct will
reflect a change in the value assigned. According to Bohrnstedt (1970), the
reliability assessments can be grouped into two major classes: measurement of
stability and measurement of equivalence. Measures of stability evaluate rela-
tionship or correlation of measures across time or evaluators, and examples
include the test–re-test method and interrater reliability. Measures of equiva-
lence evaluate relationship or correlation between two sets of instruments, and
examples include split-half, parallel form, and internal consistency methods.
Measures of stability
The test–re-test method analyzes measures obtained from the same partic-
ipants across time using the same instrument. It evaluates the stability of

measurements over time. The instrument for measurement can be any
equipment such as a weighing scale or a survey instrument with a series
of questions. For example, weight measured using a weighing scale at one
time can be correlated with weight obtained after 1 hour. The time differ-
ence between the measurements is dependent on the type of measure. A
longer time interval can influence the reliability for some measures due to
changes in the underlying construct such as a person’s weight. For other
measures involving examinations, shorter time can influence the reliability
due to knowledge of the previous test. The general rule is that the time
interval should be long enough that respondents do not remember their
responses without the change in the underlying construct. The correlation
between the two measures on the same participants provides the correla-
tion coefficient. This method is often used to ascertain the reliability of
physiological measurements such as blood glucose and blood pressure. In
fact, diagnostic criteria for a disease are often based on the test–re-test
method. Pearson’s correlation is used to calculate the correlation coefficient
for interval–ratio measures, whereas Spearman’s rank correlation is used
for ordinal measures. In general, the correlation coefficient of >0.80 is
considered as a reliable determinant (Nunnally and Bernstein 1994).
Disadvantages of the test–re-test method include inconvenience and reac-
tivity. Due to multiple measurement processes it is inconvenient to partic-
ipants and researchers. Reactivity refers to change in the underlying
construct due to testing. For example, patients responding to a question
related to compliance are sensitized to the issue of compliance and thus
provide responses that might reflect an improvement in compliance.
Interrater reliability involves analysis of measures obtained from the same
participants by different evaluators using the same instrument. For example,
pharmacists’ counseling time can be measured by two independent observers.
The reliability of such measures can be evaluated by correlating the time
measures obtained from the two observers. Cohen’s k coefficient is used for
reliability involving two evaluators, whereas Fleiss’s k is used for measures
involving multiple evaluators (Landis and Koch 1977). These measures are
calculated based on the difference between percentage of agreement among
evaluators and probability of chance agreement.
Measures of equivalence
The split-half method is one of the earliest measures of equivalence to deter-
mine questionnaire reliability. This method involves dividing the number of
items or individual questions of a survey instrument into two equivalent
halves; the correlation between the two halves provides the correlation coef-
ficient. There are two options for dividing the items or questions: one method
involves dividing the items into even and odd questions; the other involves
dividing the items into first and second halves. The decisions on the type of

splitting are based on practical considerations and type of items in the survey
instrument. Irrespective of the approach, the split-half method provides two
measures on each participant based on two equivalent-form measures, and the
measures from the halves are then used to calculate the correlation coefficient.
The basic underlying principle in the split-half method is that the two halves
are designed to measure the same underlying construct. The strength of the
split-half method is that it overcomes the problems of test–re-test methods
such as reactivity and inconvenience. The weakness of the split-half method is
that two halves must measure the same underlying construct and the reliabil-
ity coefficient can vary based on the approach used to divide the items. The
Spearman–Brown prophecy formula is generally used to obtain the correla-
tion coefficient between the split halves (DeVellis 1991). The formula is based
on the correlation coefficient between the split halves and correction needed
to divide the items in the survey instrument into two halves.
The parallel-form method is an extension of the split-half method in which
two parallel questionnaires are administered to each participant consecu-
tively. The time interval between administration of the two questionnaires
should be minimal to optimize changes in the underlying constructs. The
scores from the two forms or questionnaires are used to calculate the corre-
lation coefficient. Similar to the split-half method, the two forms should be
equivalent and measure the same underlying construct but should not be
identical. This method also overcomes the problems of test–re-test methods
such as reactivity and inconvenience. However, it may be cumbersome to the
respondents to complete two parallel questionnaires. In general it is easier
to develop similar items in the split-half method than to create parallel forms.
Consequently, the parallel-form method is seldom used in health services
research but often used in educational research involving examinations due
to experience in creating parallel exams and availability of a large pool of
questions.
The internal consistency method is the most frequently used method in
health services research. It involves correlation among all items or questions in
a questionnaire without the need to divide items or create forms. The internal
consistency method evaluates whether all items in a questionnaire are mea-
suring the same construct. It also overcomes problems associated with split-
half and parallel-form methods, such as varying reliability due to the process
used to divide the items or create the forms. Internal consistency is based on
the concept that items or questions designed to measure the same underlying
construct should be highly correlated. This means that each item or question is
used to compare consistency of responses with other items in the question-
naire for the study sample. As a result, the correlation coefficient is sample
specific. There is a need to assess the reliability of the survey instrument with a
change in the study sample. There are several ways to compute a reliability
coefficient based on the internal consistency approach. Cronbach’s a is used

to calculate the internal consistency of measures based on continuous mea-
sures (Bohrnstedt 1970; DeVellis 1991). The reliability coefficient increases
with an increase in the number of items and inter-item correlations. The
Kuder–Richardson coefficient (KR20 or KR21) is used to calculate reliability
for nominal measures. These coefficients are calculated based on the propor-
tion of same responses for an item. High conformity of responses leads to high
KR20 or a reliability coefficient of nominal measures. Cronbach’s a is an
extension of KR20 and both calculations are based on classic test theory.
Most statistical packages can calculate these reliability coefficients.
Validity
Validity, as mentioned earlier, ensures that the instrument developed for
measurement purposes truly represents the underlying construct (DeVellis
1991; Nunnally and Bernstein 1994; Trochim 2001). In addition to appro-
priateness, instruments are increasingly being evaluated for meaningfulness
and usefulness, in recent years, in order to strengthen the validity. Although
methods to ascertain validity have changed over the years, construct validity
has remained the cornerstone of all types of validity assessments. Construct
validity refers to the degree to which the instrument measures the underlying
construct. The evidence to strengthen the construct validity is based on inter-
nal structure and external relationships. The internal structure evaluates
the interrelationship of the items and underlying dimension of the construct.
The external relationship evaluates the relationship of the instrument to other
constructs.
The internal structure of the instrument should be consistent with the
underlying dimensions of the construct. Factor analysis is commonly used
in the development process of a survey instrument (DeVellis 1991; Nunnally
and Bernstein 1994). Exploratory factor analysis helps to identify various
factors or dimensions that are represented in the instrument. It also groups
items that belong together representing the underlying construct. For exam-
ple, this can be used to establish the dimensions of a quality-of-life scale.
Exploratory factor analysis is usually followed by confirmatory factor anal-
ysis to determine the extent to which statistical validity is based on the
underlying theoretical model. Although factor analysis approaches are useful,
the internal structure should not be the only basis of construct validity. The
relationship to other constructs also ascertains validity of the construct.
External relationships should be empirically tested for hypotheses developed
based on the theoretical relationships. This addresses the evidence of pre-
dictability and discernibility.
Trochim (2001) proposed that all types of validity testing methods should
strengthen the construct validity of an instrument. This can be achieved using

translational and criterion validity. The translational validity addresses the
translational or transformational aspect of construct validity, which includes
face and content validity. These validity analyses are designed to ensure that
the items in the instrument reflect the underlying construct. Face and content
validity will ensure that the items represent the intended factors or dimensions
of the construct, which can also be confirmed using exploratory factor anal-
ysis. Criterion validity refers to the relationship aspect of construct validity,
and includes concurrent, predictive, convergent, and discriminant validity.
Methods for criterion validity empirically test for theoretical relationships.
The translational validity and the criterion validity ensure appropriateness,
meaningfulness, and usefulness of the instrument. Multiple methods are
needed to strengthen the construct validity of the instrument.
Translational validity
Face validity is the simplest method to ensure translational validity. It
addresses the question: Do these items and the overall instrument measure
the underlying construct? This involves a judgment or an opinion of a layper-
son or an expert. A layperson will provide his or her perspective, mainly to
address the issues related to the interpretation of items and administration of
the instrument. Experts can provide detailed opinion about the appropriate-
ness and wording of items, organization of items, and the overall look of the
instrument with respect to the underlying construct. It primarily involves
qualitative and subjective assessment of the instrument. Consequently, it is
considered as the weakest form of validity assessment. It is often used in the
development process to refine an instrument.
Content validity refers to the representative nature of the items to capture
the underlying dimension of the construct. It presents the relationship of items
to the dimensions of the underlying construct. The content validation process
requires a clear definition of the dimensions of the underlying construct and
ways to ensure that the selected items represent the relevant dimensions. The
dimensions of the construct can be defined from the literature or expert
opinion. For example, measurement of healthcare quality requires items or
questions related to the structure of the healthcare system, process of obtain-
ing healthcare, and outcomes of healthcare obtained (Donabedian 2003). A
content expert’s opinion can be sought to evaluate whether the measurement
items represent the defined dimensions. Although there is some subjectivity in
the process, seeking the opinions of multiple experts can reduce subjectivity
and improve the face validity. Also, analytical measures such as content
validation ratio, content validation form, and content validation index can
strengthen the content validity (Shultz and Whitney 2005). Content validity
will ensure that the items and the overall instrument reflect the dimensions of
the construct.

Criterion validity
Criterion validity addresses the relationship aspect of the construct validity by
attesting to the relationships between the instrument and criterion, or other
measures, based on theory and practice. Selection of the criterion plays an
important role in criterion validity. The evidence and the extent of the inter-
relationship of criteria strengthen or weaken the construct validity. A strong
relationship means that the criterion is well validated and accepted. Often the
criterion selected is external and considered the “gold standard.” The theo-
retical and practical knowledge about the issues are critical in selection of the
criterion. For example, compliance measured using self-reports or refill his-
tory can be validated using the criterion of drug levels in blood or urine
because these are considered the gold standard. As mentioned earlier, types
of criterion validity include concurrent, predictive, convergent, and discrim-
inant validity.
Concurrent validity is a type of criterion validity that refers to the rela-
tionship between the instrument and the criterion measured at the same point
in time. The criterion selected for concurrent validity should measure the same
underlying construct as that of the instrument. The selected criterion should
be a standard measure. The rationale for concurrent validity is that, if
the instrument and the criterion are administered at the same time and mea-
sure the same underlying construct, then there should be strong correlation
between the two measures. For example, compliance measured using self-
reports can be validated by comparing responses with the drug levels in blood
or urine measured at the same time.
Predictive validity is a type of criterion validity that addresses the relation-
ship between the instrument and criterion measured at a future time. The
criterion and the instrument are not measuring the same underlying construct
as in the concurrent validity. However, the instrument should be able to
predict the criterion. For example, compliance measures based on refill his-
tory have been shown to predict the health expenditure in patients with
diabetes. This is based on the hypothesis that disease is managed better in
compliant than in noncompliant patients, and thus leads to a decrease in
healthcare expenditure. The rationale for predictive validity is that, if the
patients with diabetes are compliant, they will incur less expenditure due to
better disease-state management. Therefore, the criterion selected for predic-
tive validity should be based on theory and practice.
Convergent and discriminant validity are two sides of the same concept.
Convergent validity refers to convergence or a strong relationship between the
instrument and the criterion, which are theoretically similar. Discriminant
validity refers to little or no relationship between the instrument and the
criterion, which are theoretically different. Convergence validity is similar
to concurrent validity but it is not restrictive with respect to time of

administration. The concept of convergent and discriminant validity is based
on the principle that, if the instrument is valid, it will be strongly correlated
with measures that are similar and will not be associated with measures that
are dissimilar. For example, instruments to measure pain and overall quality
of life will be strongly associated because they measure similar concepts.
Conversely, pain measures are less likely to be associated with perceptions
on economy as these are dissimilar concepts.
Measurement process and practice
Understanding the measurement theory and practice is vital in conducting
empirical research. An existing knowledge base defines the methodology
to operationalize a construct, which includes identification of acceptable
behavior specimens, data collection, and assignment of values to develop the
variable for an underlying construct. Some constructs are easy to measure,
such as expenditure; other constructs, such as quality, need a strong under-
standing of underlying theory in order to operationalize. Prescription expen-
ditures are usually captured using secondary data sources such as claims
data. The behavior specimen is reflected in claims data and it captures
payment by insurance companies and other sources. The data collection
involves use of secondary data and value assignment of expenditures
involves a ratio scale. Other constructs such as quality are complex and
require significant effort. The existing knowledge base suggests that quality
measurements should be based on the underlying dimensions. For example,
quality of medical care is based on measures of structure, process, and
outcome (Donabedian 2003). The operational definitions and measure-
ments processes for each of these dimensions are different. According
to Donabedian (2003), measures of quality are relevant only when there is
an interrelationship for structure, process, and outcome. Consequently,
greater understanding of the underlying theory and advances in data collec-
tion methodologies play an important role in the measurement process and
operationalization of healthcare constructs.
Reliability and validity issues are critical in the identification of behavior
specimen and data collection phases. Concepts of reliability and validity are
valuable to develop and improve a strong research instrument. Accordingly,
they should be considered tools for continuous quality improvement of
the measurement process. A well-defined construct helps to ensure the con-
struct validity of an instrument, which includes translational and criterion
validity. Translational validity includes face and content validity. The con-
siderations of face and content validity can be incorporated in the identifica-
tion of the behavior specimen phase to ensure that instrument development is
consistent with the underlying dimensions of the construct. The criterion

validity is difficult to incorporate into the behavior specification phase. In
general the considerations of translational validity are likely to ensure crite-
rion validity. Criterion validity can be tested only by correlating measures
from the instrument with a criterion. The test findings will reveal needed
improvements in the development phase. Figure 4.1 provides a schematic
diagram for the measurement process and practice.
In the data collection phase, the developed instrument is utilized to collect
the data. Various research techniques can be employed to ensure reliability
and validity of data collection. Standardized data collection and administra-
tion methods can minimize random errors and improve reliability. Pre-testing
of the instrument can help to identify items that require clarification. It can
also help to improve the organization of the instrument. Response biases can
be minimized using pre-tested items, techniques of blinding, utilization of
trained interviewers, and consistent data collection methods. The data col-
lection process should minimize random errors and control for nonrandom
errors to maximize reliability and validity. The tests for reliability and validity
will ensure that the instrument developed and utilized for research is reliable
and valid.
Summary and conclusions
The measurement process is designed to record and capture the underlying
construct. This involves identification of acceptable specimens, data
1. Identification of acceptable specimens
2. Data collection
3. Assignment of values
Reliability testing
Validity testing
Final variable
Construct
Figure 4.1 Measurement process and practice.

collection of specimens, and conversion of specimens to a quantitative
variable. The decisions made at each of these interrelated steps are based
on the existing knowledge base of the construct. Reliability and validity
issues are critical in the measurement process. Reliability addresses stability
and equivalence of the measurement process. The tests of translational and
criterion validity are designed to ensure construct validity of the measure-
ment process. Construct validity addresses the extent to which the variable
measures the underlying construct. A reliable and valid measurement pro-
cess will minimize measurement errors and thereby strengthen the research.
Measurement also forms the basis of subsequent steps in research such as
statistical analysis.
Review topics
1 Discuss levels of measurement using examples.
2 Describe the concept of reliability and methods to evaluate reliability.
3 Discuss common types of measurement errors.
4 Describe the concept of validity and methods to evaluate validity.
5 Describe the measurement process using an example in pharmaceutical
practice and policy research.
References
Bohrnstedt GW (1970). Reliability and validity assessment in attitude measurement. In: Summers
GF (ed.), Attitude Measurement. Chicago, IL: Rand McNally, 80–99.
DeVellis RF (1991). Scale Development: Theory and applications. Thousand Oaks, CA: Sage
Publications.
Donabedian A (2003). An Introduction to Quality Assurance in Health Care. New York: Oxford
University Press.
Farmer KC (1999). Methods for measuring and monitoring medication regimen adherence in
clinical trials and clinical practice. Clin Ther 21: 1074–90.
Hayes RB, Taylor DW, Sackett DL (1979). Compliance in Health Care. Baltimore, MD: Johns
Hopkins University Press.
Landis JR, Koch GG (1977). The measurement of observer agreement for categorical data.
Biometrics 33: 159–74.
Nunnally JC, Bernstein IH (1994). Psychometric Theory, 3rd edn. New York: McGraw-Hill.
Shultz KS, Whitney DJ (2005). Measurement Theory in Action: Case studies and exercises.
Thousand Oaks, CA: Sage Publications.
Stevens SS (1946). On the theory of scales of measurement. Science 103: 677–80.
Summers GF (1970). Introduction. In: Summers GF (ed.), Attitude Measurement. Chicago, IL:
Rand McNally, 1–21.
Trochim WMK (2001). The Research Methods Knowledge Base, 2nd edn. Cincinnati, IL: Atomic
Dog Publishing.
Viswanathan M (2005). Measurement Error and Research Design. Thousand Oaks, CA: Sage
Publications.

Online resources
Agency for Healthcare Research and Quality (AHRQ). “National Quality Measures
Clearinghouse.” Available at: www.qualitymeasures.ahrq.gov.
Centers for Disease Control and Prevention (CDC). National Center for Health Statistics. Series 2.
“Data Evaluation and Methods Research.” Available at: www.cdc.gov/nchs/products/series.
htm#sr2.
Centers for Disease Control and Prevention (CDC). National Center for Health Statistics. Series 6.
“Cognition and Survey Measurement.” Available at: www.cdc.gov/nchs/products/series.
htm#sr6.
The Leapfrog Group. Available at: www.leapfroggroup.org.

Rmppp ch04web

More Related Content

What's hot (20)

Viewers also liked (15)

Similar to Rmppp ch04web (20)

Recently uploaded (20)

Rmppp ch04web