Professional Documents
Culture Documents
Radiology
318
that are either primarily educational or leading results, and inappropriate patient have a different value in different people,
have links to other resources such as sta- care. The benefits of statistical software in different places, or at different times.
tistical software. In addition, some sug- generally far outweigh the costs, but Such variables are often referred to as
gested additional readings are listed in proper measurement, study design, and random variables when the value of a
Appendix C. good judgment should prevail over the particular outcome is determined by
ease with which many analyses can be chance (ie, by means of random sam-
Radiology
Precedence Does the independent variable vary before the effect exhibited in the
equal (eg, temperature in degrees Celsius, dependent variable?
T scores reported for bone mineral den- Nonspuriousness Can the empirical correlation between two variables be explained
away by the influence of a third variable?
sity). Ratio data are the same as interval
Plausibility Is the expected outcome biologically plausible and consistent with
data but have a true zero starting point. theory, prior knowledge, and results of other studies?
As noted in the examples above, the val-
ues of degrees Celsius and T score can
take on both positive and negative num-
bers. Examples of ratio data would be
heart rate, percentage vessel stenosis, and and subsequent malignancy). Content plore (not prove) connections between
respirations per minute. Interval and ra- validity is the extent to which the indi- independent and dependent variables. A
tio data are also referred to as quantita- cator reflects the full domain of interest dependent variable (sometimes called
tive variables, since they have a direct (eg, tumor shrinkage may be indicated by the response variable) is a variable that
numeric interpretation. In most analyses, tumor width, height, or both). Construct contains variations for which we seek an
it does not matter whether the data are validity is the degree to which one mea- explanation. An independent variable is
interval or ratio data. sure correlates with other measures of the a variable that is thought to affect (cause)
same concept (eg, does a positive MR changes in the dependent variable. Cau-
Continuous and Discrete Variables study for multiple sclerosis correlate with sation is implied when statistically signif-
physical examination findings, patient icant associations are found between an
Variables such as weight and diameter symptoms, or laboratory results?). Face independent and a dependent variable,
are measured on a continuous scale, validity evaluates whether the indicator but causation can never be truly proved.
meaning they can take on any value appears to measure the concept. As an Proof is always an exercise in logical de-
within a given interval or set of intervals. example, it is unlikely that an MR study duction tempered with a degree of uncer-
As a general rule of thumb, if a subdivi- of the lumbar spine will facilitate a diag- tainty (18,19), even in experimental de-
sion between intervals makes sense, the nosis for lost memory and disorientation. signs (such as randomized controlled
data are continuous. As an example, a trials).
time interval of minutes can be further Association Statistical techniques provide evidence
divided into seconds, milliseconds, and that a relationship exists between inde-
an infinite number of additional frac- The connection between variables is
pendent and dependent variables through
tions. In contrast, discrete variables such often referred to as association. Associa-
the use of significance testing and mea-
as sex, the five-point BI-RADS final assess- tion, also known as covariation, is exhib-
sures of the strength of association. This
ment scale, race, and number of children ited by measurable changes in one vari-
evidence must be supported by the theo-
in a household have basic units of mea- able that occur concurrently with
retical basis and logic of the research. The
surement that cannot be divided (one changes in another variable. A positive
Table presents a condensed list of ele-
cannot have 1.5 children). association is represented by changes in
ments necessary for a claim of causation.
the same direction (eg, heart rate in-
The first attempt to provide an epidemi-
Reliability and Validity creases as physical activity increases).
ologic method for evaluating causation
Negative association is represented by
Measurement accuracy is directly re- was performed by A. G. Hill and adapted
concurrent changes in opposite direc-
lated to reliability and validity. Reliabil- for the well-known U.S. Surgeon Gener-
tions (hours per week spent exercising
ity is the extent to which the repeated use al’s report, Smoking and Health (1964)
and percentage body fat). Spurious asso-
of a measure yields the same values when (18,19). The elements described in the
ciations are associations between two
no change has occurred. Therefore, reli- Table serve to remind us that causation is
variables that can be better explained by
ability can be evaluated empirically. Poor neither a simple exercise nor a direct
a third variable. As an example, if after
reliability negatively affects all studies. As product of statistical significance. This is
taking medication for a common cold for
an example, reliability can depend on why many believe the optimal research
10 days the symptoms disappear, one
who performs the measurement and technique to establish causation is to use
could assume that the medication cured
when, where, how, and from whom the a randomized controlled experiment.
the illness. Most of us, however, would
data are collected.
probably agree that the change is better
Validity is the extent to which a mea-
explained in terms of the normal time
sure is an accurate representation of the MAINTAINING PERSPECTIVE
course of a common cold rather than a
concept it is intended to operationalize.
pharmacologic effect.
Validity cannot be confirmed empiri- Rothman and Greenland (19) wrote,
cally—it will always be in question. Al- “The tentativeness of our knowledge
Causation
though there are several different con- does not prevent practical applications,
ceptualizations of validity, the following There is a difference between the deter- but it should keep us skeptical and criti-
provides a brief overview. Predictive va- mination of association and that of cau- cal, not only of everyone else’s work but
lidity refers to the ability of an indicator sation. Causation cannot be proved with of our own as well.”
to correctly predict (or correlate with) an statistics. With this caveat in mind, sta- A basic understanding of measurement
outcome (eg, imaged abnormal lesion tistical techniques are best used to ex- will enable radiologists to better under-
sufficient insight to recognize that statis- a measure that is represented by a limited The Little Handbook of Statistical Prac-
number of values. tice: www.tufts.edu/⬃gdallal/LHSP.HTM
tical sophistication and significance test-
Face validity.—Face validity evaluates Rice Virtual Lab in Statistics: www.ruf.rice
ing are tools, not ends in themselves. Sta-
whether the indicator appears to measure .edu/⬃lane/rvls.html
tistical techniques, however, are useful in
the abstract concept. Concepts and Applications of Inferential
providing summary measures of con-
Independent variable.—The independent Statistics: faculty.vassar.edu/⬃lowry/webtext
cepts and helping researchers decide, variable can be manipulated to affect varia- .html
given certain assumptions, what is mean- tions or responses in another variable. StatSoft Electronic Textbook: www.statsoft
ingful in a statistical sense (more about Interval data.—These variables classify ob- .com/textbook/stathome.html
this in future articles). As new techniques jects according to type and logical order but Hypertext Intro Stat Textbook: www
are presented in this series, readers also require that differences between levels .stat.ucla.edu/textbook/
should remind themselves that statistical of a category are equal. Introductory Statistics: Concepts, Models,
significance is meaningless without clin- Nominal data.—These are variables that and Applications: www.psychstat.smsu.edu/
ical significance. classify objects according to type or charac- sbk00.htm
teristic. Statnotes: An Online Textbook: www2
Operationalize.—This is the process of cre- .chass.ncsu.edu/garson/pa765/statnote.htm
WHAT COMES NEXT ating a measure of an abstract concept. Research Methods Knowledge Base: trochim
Ordinal data.—These are variables that .human.cornell.edu/kb/
This introduction to measurement will classify objects according to type or kind The following is a list Web links to statis-
be followed by a series of articles on basic but also have some logical order. tical software (accessed May 14, 2001).
biostatistics. The series will cover topics Predictive validity.—This is the ability of Stata Software (links to software provid-
on descriptive statistics, probability, sta- an indicator to correctly predict (or corre- ers): www.stata.com/links/stat_software.html
tistical estimation and hypothesis test- late with) an outcome. EpiInfo, free software downloads avail-
ing, sample size, and power. There will Random variable.—This type of variable is a able from the Centers for Disease Control
also be more advanced topics introduced, measure where any particular value is based and Prevention: www.cdc.gov/epiinfo/
such as correlation, regression modeling, on chance by means of random sampling.
statistical agreement, measures of risk Ratio data.—These variables have a zero APPENDIX C: SUGGESTED
and accuracy, technology assessment, re- starting point and classify objects according
GENERAL READINGS
ceiver operating characteristic curves, to type and logical order but also require
and bias. Each article will be written by that differences between levels of a category
The following is a list of suggested read-
experienced researchers using radiologic be equal.
ings.
Reliability.—Reliability is the extent to
examples to present a nontechnical ex- Pagano M, Gauvreau K. Principles of bio-
which the repeated use of a measure yields
planation of a statistical topic. statistics. Belmont, Calif: Duxbury, 1993.
the same value when no change has occurred.
Motulsky H. Intuitive biostatistics. New
Spurious association.—This is an association
York, NY: Oxford University Press, 1995.
APPENDIX A: KEY TERMS between two variables that can be better ex-
Rothman KJ, Greenland S, eds. Modern
plained by or depends greatly on a third vari-
epidemiology. Philadelphia, Pa: Lippincott-
Below is a list of the common terms and able.
Raven, 1998.
definitions related to measurement. Statistical control.—This refers to holding
Gordis L, ed. Epidemiology. Philadelphia,
Abstract concept.—The starting point for the value of one variable constant in order
Pa: Saunders, 1996.
measurement, an abstract concept is best to clarify associations among other vari-
Oxman AD, Sackett DL, Guyatt GH. Us-
understood as a general idea in linguistic ables.
ers’ guides to the medical literature. I. How
form that helps us describe reality. Statistical inference.—This is the process
to get started. The Evidence-Based Medicine
Association.—An association is a measur- whereby one reaches a conclusion about a
Working Group. JAMA 1993; 270:2093–2095.
able change in one variable that occurs con- population on the basis of information ob-
[This is from an ongoing series through year
currently with changes in another variable. tained from a sample drawn from that pop-
2000.]
Positive association is represented by change ulation. There are two such methods, statis-
in the same direction. Negative association is tical estimation and hypothesis testing. References
represented by concurrent changes in oppo- Validity.—Validity is the extent to which 1. Disraeli B. Quoted by: Twain M. An auto-
site directions. a measure accurately represents the abstract biography of Mark Twain. Neider C, ed.
Constant.—A constant is an attribute of a concept it is intended to operationalize. New York, NY: Columbia University
concept that does not vary. Variable.—A variable is a measure of a Press, 1995; chapter 29.
concept that can take on more than one 2. Altman DG, Bland JM. Improving doc-
Construct validity.—Construct validity is tor’s understanding of statistics. J R Stat
the degree to which one measure correlates value from one observation to the next. Soc 1991; 154:223–267.
with other measures of the same abstract 3. Hillman BJ, Putnam CE. Fostering re-
concept. APPENDIX B: WEB search by radiologists: recommendations
Content validity.—Content validity is the RESOURCES of the 1991 summit meeting. Radiology
1992; 182:315–318.
extent to which the indicator reflects the
4. Doubilet PM. Statistical techniques for
full domain of interest. The following is a list of links to general medical decision making: application to
Continuous variable.—This type of variable statistics resources available on the Web (ac- diagnostic radiology. AJR Am J Roentge-
is a measure that can take on any value cessed May 14, 2001). nol 1988; 150:745–750.