100 .Chapter4: ResearchDesign
Faulty Reasoningabout Unitsofllnalysis:
TheEcologicalFallacyandReductionism
At this point, it's appropriateto introduce two types
of faulty reasoningthat you should be aware of: the
ecologicalfallacy and reductionism.Each repre-
sents a potentialpitfall regardingunits of analysis,
either of which can occur in doingresearch and
drawing conclusionsfrom the results.
TheEcologicalFallacy
In this context, "ecological" refers to groups or
sets or systems: somethinglarger than individuals.
The ecologicalfallacy is the assumptionthat
somethinglearned about an ecologicalunit says
something about the individuals making up that
unit. Let's consider a hypotheticalillustration of this
fallacy.
Supposewe are interestedin learning about the
nature of electoralsupport received by a femalepo-
litical candidateina recent citywide election.Let's
assumethat we have the vote tally for each precinct
sothat we can tell which precincts gave her the
greatestsupport and which the least. Assume also
that we have census data describing some charac-
teristicsof these precincts. Our analysis of such data
might showthat precincts with relatively young
voters gave the female candidate a greaterpropor-
tion of their votes than did precincts with older vot-
ers. We might be tempted to conclude from these
findingsthat youngvoters are more likely to vote
for female candidatesthan are older voters-in
other words, that age affects support for women in
politics.In reachingsuch a conclusion,we run the
risk of committingthe ecologicalfallacybecause it
may have been the older voters in those "yo~~lg"
precinctswho voted for the woman. Our problem
is that we have examinedprecincts as our units of
analysisbut wish to draw conclusionsabout voters.
The sameproblem would arise if we discovered
that crimerates were higher in cities having large
African-Americanpopulationsthan in those with
few African Americans.We would not know if the
crimeswere actually committedby African Ameri-
cans. Or if we found suiciderates higher in Protes-
tant countriesthan in Catholic ones, we stillcould
not know for sure that more Protestants than
Catholics committedsuicide.
In spite of these hazards, socialscientists very
often have little choicebut to address a particular
research question through an ecological analysis.
Perhapsthe most appropriatedata are simply not
available.For example, the precinct vote tallies and
the precinct characteristicsmentioned in our initial
example might be easy to obtain, but we may not
'
have the resourcesto conduct a postelection survey
of individualvoters. In such cases, we may reach a
tentative conclusion, recognizing and noting the
risk of an ecologicalfallacy.
While you shouldbe carefulnot to committhe
ecological fallacy, don't let these warnings lead you
into committingwhat we might callthe individual-
isticfallacy. Somepeople who approachsocialre-
search forthe first time have trouble reconciling
general patterns of attitudesand actionswith indi-
vidual exceptions.As we discussedin Chapter 2,
generalizationsand probabilistic statementsare not
invalidatedby individual exceptions.Your knowing
a rich Democrat, for example, doesn't deny the fact
that most rich people vote Republican-as a gen-
eral pattern. Similarly, if you know someonewho
has gotten rich without any formal education,that
doesn't deny the generalpattern of higher educa-
tion relatingto higher income.
The ecologicalfallacy dealswith somethingelse
altogether-confusing units of analysis in such a
way that we draw conclusions about individuals
based solelyon the observationof groups. Although
the patterns observedbetween variables at the level
of groupsmay be genuine,the danger lies in rea-
soning from the observed attributesof groupsto
the attributesof the individualswho made up
those groupswhen we have not actually observed
individuals.
Reductionism
A secondtype of potentially faulty reasoningre-
latedto units of analysis is reductionism. Reduc-
tionismmeans seeingand explaining complexphe-
nomena in terms of a single,narrow concept or set
of concepts. Thus, we "reduce" what in reality is
complexto a simple explanation.
For instance, scientists from different disciplines
tend to look at differenttypes of answers and ig-
nore the others. Sociologists tend to consider only
sociologicalvariables (suchas values, norms, and
roles),economistsonly economicvariables (suchas
supply and demand, marginal value), and psychol-
ogists only psychologicalvariables (suchas person-
ality types, traumas).Explainingall or most human
behavior in terms of economicfactorsis called eco-
nomic reductionism; explaining all or most human
behavior in terms of psychological factorsis called
psychologicalreductionism; and so forth. Notice
how this issue relates to the discussion of theoreti-
calparadigmsin Chapter2.
In another example, supposewe ask what
causedthe AmericanRevolution.Was it a shared
commitment to the value of individual liberty?
The economicplight of the colonies in relation to
Britain?The megalomania of the founding fathers?
As soon as we inquire about the single cause, we
run the rislc of reductionism.
Reductionismof any type tends to suggestthat
particular units of analysis or variables are more
relevant than others.If we were to regard shared
values as the cause of the American Revolution,
our unit of analysiswould be the individual colo-
nist. An economist,though, might choose the
13colonies as units of analysis and examine the
economicorganizationsand conditions of each.
Apsychologistmight choose individual leaders as
the units of analysis for purposes of examining
their personalities.
Like the ecologicalfallacy,reductionism can
occur when we use inappropriateunits of analysis.
The appropriateunit of analysis for a given re-
search question, however, is not always clear. So-
cial scientists, especially across disciplinarybound-
aries, often debatethis issue.
The Time Dimension
Sofar in this chapter, we have regarded research
design as a process for deciding what aspects we
shall observe, of whom, and for what purpose.
Now we must consider a set of time-relatedoptions
TheTimeDimension .101
that cuts across each of these earlier considerations.
We can cllooseto make observationsmore or less
at one time or over a longperiod.
Time plays many roles in the design and execu-
tion of research, quite aside from the time it takes
to do research. Chapter 3 noted that the time se-
quence of events and situationsis critical to deter-
mining causation (apoint we'll return to in Part 4).
Time also affectsthe generalizabilityof research
hdings. Do the descriptions and explanations re-
sultingfrom a particular study accuratelyrepresent
the situation of ten years ago, ten years from now,
or only the present?Researchers have two princi-
pal optionsavailable to deal with the issue of time
in the design of their research: cross-sectionalstud-
ies and longitudinal studies.
Cross-SectionalStudies
A cross-sectionalstudyinvolves observationsof
a sample, or cross section, of a population or phe-
nomenon that are made at one point in time. Ex-
pioratory and descriptive studies are often cross-
sectional.A single U.S. Census, for instance,is a
study aimed at describingthe U.S. population at a
given time.
Many explanatorystudies are also cross-
sectional.A researcher conducting a large-scale
nationalsurvey to examinethe sources of racial
and religiousprejudice would, in alllikelihood,be
dealing with a singletime frame-taking a snap-
shot, so to speak, ofthe sources ofprejudice at a
particular point in history.
Explanatory cross-sectionalstudieshave an in-
herent problem.Although their conclusionsare
based on observationsmade at only one time, typi-
cally they aim at understandingcausal processes
that occur over time.This problemis somewhat
akin to that of determiningthe speed of a moving
object on the basis of a high-speed,stillphotograph
that freezesthe movement of the object.
Yanjie Bian, for example, conducteda survey of
workers in Tianjin, China, forthe purpose of study-
ing stratificationin contemporary,urban Chinese
society. In undertaking the surveyin 1988, how-
ever, he was conscious ofthe important changes
Určeno pouze pro studijní účely
102 .Chapter4: ResearchDesign
brought about by a series of national campaigns,
such as the Great Proletarian CulturalRevolution,
datingfrom the ChineseRevolution of 1949 (which
brought the Chinese Communistsinto power) and
continuinginto the present.
These campaigns altered political atmospheres
and affectedpeople's worlc and nonwork activi-
ties. Because of these campaigns, it is difficultto
draw conclusionsfrom a aoss-sectional social
survey,such as the one presented in this book,
about general patterns of Chinese workplaces
and their effectson workers. Such conclusions
may be limited to one period of time and are
subject to further tests based on data collected
at other times.
(1774:17)
The problem of generalizations about social life
from a "snapshotr'is one this book repeatedly ad-
dresses. One solution is suggestedby Bian's find
comment-about data collected "at other times":
Socialresearch ofteninvolves revisitingphenom-
ena andbuilding on the results of earlier research.
longitudinalStudies
In contrastto aoss-sectionalstudies, a longitudi-
nal studyis designed to permit observationsof the
same phenomena over an extended period. For ex-
ample, a researcher can participatein and observe
the activitiesof a UFO cult from its inceptionto its
demise. Other longitudinal studiesuse records or
artifactsto study changes over time. Inanalyses of
trends in newspaper editorials or Supreme Court
decisions over time, for example, the studies are
longitudinalwhether the researcher's actual obser-
vations and analysesare made at one time or over
the course of the actual events under study.
Many field research projects, involving direct
observation and perhaps in-depth interviews,are
n a N d y longitudinal. For example,when Ramona
Asher and GaryFine (1991)studiedthe life experi-
ences of the wives of alcoholicmen, they were in a
position to examine the evolutionof the women's
troubled marital relationshipsover time, sometimes
even including the reactions of the subjectsto the
research itself.
In a classic study, Wzeln Proplzecy Fails (1956),
Leon Festinger,Henry Reicker, and Stanley
Schachterwere specificallyinterested in learning
what happened to a flying saucer cult when the
cult's predictions of an alien encounter failedto
come true. Would the cult members close down the
group, or would they become all the more commit-
ted to their beliefs? A longitudinalstudy was re-
quired to provide an answer. (Theyredoubled their
effortsto get new members.)
Longitudinal studies can be more difficult for
quantitativestudies such as large-scalesurveys.
Nonetheless,they are often the best way to study
changes over time. There are three specialtypes of
longitudinalstudiesthat you should know about:
trend studies, cohort studies, and panel studies.
TrendStudies
A trend studyis a type of longitudinalstudy that
examineschanges within a population over time.
A simple example is a comparison of U.S. Censuses
over a period of decades, showing shiftsin the
makeup of the national population.A similaruse
of archivaldata was made by Michael Carpini and
ScottIceeter (1991),who wanted to know whether
contemporaryU.S. citizens were better or more
poorly informed about politics than were citizens of
an earlier generation.To find out, they compared
the results of several Gallup Polls conducted dur-
ing the 1940sand 1950swith a 1989survey that
asked several of the same questionstapping politi-
calknowledge.
Overall, the analysis suggestedthat contempo-
rary citizens were slightlybetter informed than
were earlier generations.In 1989, 74percent of the
sample couldname the vice president of the United
States,comparedwith 67 percent in 1952. Sub-
stantiallyhigher percentages could explainpresi-
dentialvetoes and congressionaloverrides of vetoes
than couldpeople in 1947.On the other hand,
more of the 1947sample couldidentifytheir U.S.
representative (38percent) than could the 1989
sample (29percent).
An in-depth analysis, however, indicates that
the slight increase in politicalknowledge resulted
from the fact that the people in the 1989sample
were more highly educated than were those from
TheTime Dimension . 103
TABLE 4-1
Ageand PoliticalLiberalism
1972 1977 1982 1987.. ..
to to to to
SurveyDates 1974 1980 1984 1989
Age ofCohort 20-24 25-29 30-34 35-39
Percentwhowould let
the communistspeak 72% 68% 73% 73%
earlier samples. When educationallevelswere
taken into account, the researchers concludedthat
political knowledge had actuallydeclined within
specificeducationalgroups.
CohortStudies
In a cohort study, a researcher examines specific
subpopulations, or cohorts, as they change over
time. Typically, a cohort is an age group, such as
those people born during the 1950s,but it can also
be some other time grouping, such as people born
duringthe Vietnam War, people who got married
in 1994,and so forth. An example of a cohort study
would be a series ofnational surveys, conducted
perhaps every 20 years, to study the attitudes of the
cohort born during World War 1Ttoward U.S. in-
volvement in global affairs.A sample of people 15-
20 years of age might be surveyedin 1960, another
sample of those 35-40 years of age in 1980,and
another sample of those 55-60 years of age in
2000. Although the specificset of people studied in
each surveywould differ, each samplewould rep-
resent the cohort born between 1940and 1945.
James Davis (1992)turned to a cohort analysis
in an attempt to understand shifting political orien-
tations during the 1970sand 1980sin the United
States. Overall, he found a Liberal trend in issues
such as race, gender, religion, politics, airne, and
freespeech.But did this trend represent people in
generalgettinga bit more liberal, or did it merely
reflect more liberal younger generationsreplacing
the conservativeolder ones?
To answer this question, Davis examined na-
tional surveysconducted in four time periods, five
years apart. In each survey, he grouped the respon-
dents into age groups, also five years apart. This
strategyallowedhim to compare different age
groups at any given point in time as well as fol-
low the political developmentof each age group
over time.
One of the questions he examined was
whether a person who admitted to being a com-
munist should be allowed to speakin the respon-
dents' communities. Consistently, the younger re-
spondentsin each period of time were more willing
to let the communist speak than were the older
&es. Among those aged 20-40 in the first set of
the survey, for example, 72 percent took this liberal
position, contrastedwith 27 percent among re-
spondents 80 and older.What Davis found when
he examined the youngest cohort over time is
shown in Table 4-1.
This pattern of a slight, conservativeshift in the
1970s,followedby a liberal rebound in the 1980s,
typifies the several cohortsDavis analyzed (J.Davis
1992:269).
PanelStudies
Though similar to trend and cohort studies, a
panel study examinesthe same set of people each
time. For example,we could interview the same
sample of voters every month during an election
campaign, askingfor whom they intended to vote.
Though such a studywould allow us to analyze
overall trends in voter preferencesfor different can-
didates, it would also show the precise patterns of
persistence and changein intentions. For example,
a trend study that showedthat CandidatesA and B
eachhad exactlyhaIf of the voters on September 1
and on October 1 could indicatethat none of the
Určeno pouze pro studijní účely
104 .Chapter4: ResearchDesign TheTimeDimension .105
electoratehad changed voting plans, that all of the
voters had switched allegiance to the other candi-
date, or somethingin between. Apancl study
would eliminate this confusionby showingwhat
kinds of voters switched from A to B and what
ldnds switched from B to A, as well as other facts.
Joseph Veroff, Shirley Hatchett, and Elizabeth
Douvan (1992)wanted to learn about marital ad-
justment amongnewlyweds,looldng for differences
between white and African-Americancouples. To
get subjectsfor study, they selecteda sample of
coupleswho applied for marriage licensesin Wayne
County, Michigan, Aprilthrough June 1986.
Concerned about the possibleimpacttheir re-
searchmight have on the couples' marital adjust-
ment, the researchers dividedtheir samplein half
at random: an experimentalgroup and a control
group (conceptswe'll explore further in Chapter 8).
Couplesin the formergroup were intensivelyinter-
viewed over a four-yearperiod, whereas the latter
group was contactedonlybriefly eachyear.
By studying the same couples over time, the
researchers could followthe speczc problemsthat
arose and the way the couples dealt with them.
As a by-product of their research, they foundthat
those studiedthe most intensely seemedto achieve
a somewl~atbetter marital adjustment. The re-
searchersfelt that the interviewsmay have forced
couples to discussmatters they may have otherwise
buried.
ComparingtheThreeTypesofLongitudinal5tudies
To reinforcethe distinctionsamong trend, cohort,
and panel studies,let's contrast the three study de-
signsin terms of the samevariable:political party
affiliation.A trend study might look at shiftsin U.S.
religious af6liationsover time, as the GallupPoll
does on a regular basis.A cohort studymight fol-
low shiftsin religious affiliationsamong "the De-
pressiongeneration,"specifically, say, people who
werebetween 20 and 30 in 1932.We could study a
sample of people 30-40 years old in 1942,a new
sample of people aged 40-50 in 1952,and so forth.
Apanel study could start with a sample of the
whole population or of some special subset and
study those specificindividuals over time. Notice
that only the panel study would give a fullpicture
of the shifts ainong the various religions as well as
into or out ofthe "none" category.Cohort and
trend studieswould uncover only net changes.
Longitudinalstudiesin generalhave an obvious
advantageover cross-sectionalonesin providingin-
formationdescribing processesover time. But this
advantageoftencomes at a heavy cost in both time
and money, especiallyin a large-scalesurvey.Ob-
servationsmay have to be made at the time events
are occurring,and the method of observationmay
require many research workers.
panel studies,which offer the most compre-
hensive data on changes over time,face a special
problem:panel attrition. Some of the respondents
studiedin the fist wave of the surveymay not par-
ticipate inlater waves, whether by choice or cir-
cumstance.The danger is that those who drop out
of the studymay not be typical, thereby distort-
ing the results of the study. Thus, when Carol S.
Aneshenscl and colleagues compared Latina and
non-Latina adolescents' sexual behavior by means
of a panel study, they loolced for and found differ-
ences in characteristicsof survey dropoutsamong
Latinas born in the United Statesand those born
in Mexico.These differencesneeded to be taken
into accountto avoidmisleadingconclusionsabout
differencesbetween Latinas and non-Latinas
(Aneshenselet al. 1989).For a further comparison
of the three types of longitudinal studies,see the
box entitled "TheTimeDimensionand Aging."
Approximating 1ongitudinalStudies
Longirudinal studiesdo not always provide a fea-
sible or practical means of studyingprocesses that
take place over time.Fortunately, researchersoften
can draw approximate condusions about such
processes even when only cross-sectional data are
available.Here are some ways to do that.
Sometimes, cross-sectional data implyprocesses
over time on the basis of simplelogic.For example,
in the study of student druguse conducted at the
University of Hawaii that I mentioned in Chapter2,
studentswere asked to report whether they had
ever tried each of severalillegal dmgs.The study
byJoseph1.Leon
BehavioralScienceDepartment,California
StatePolytehnicUniversity,Pamona
O n e way to identify the type of time dimen-
sion used in a study is to imagineanumber
of different research projectson growing
older inthe American society. Ifwe studieda
sample of individualsin 1990and compared
the different age groups, the designwould
be termed cross-sectional. If we drew another
sample of individualsusingthe same study
instrumentin the year 2000 and compared
the new datawith the 1990data,the design
would betermed trend.
Supposewe wishedto study only those
individualswho were 51-60 in the year 2000
and comparethem with the 1990sample of
41-50-year old persons(the41-50 age co-
hort);this study designwould betermed co-
hort.The comparisoncould be madefor the
51-60 and 61-70 age cohorts as well. Now, if
we desired to do apanelstudy on growing
older in America, we would draw asample in
the year 1990and,using the samesampled
individuals in the year 2000,do the study
again.Remember,there would befewer
peoplein the year 2000study becauseallthe
41-50-year-old people in 1990are 51-60 and
therewould be no41-50-year-old individuals
inthe year 2000 study. Furthermore,someof
the sampled individualsin 1990would no
longerbe alive in the year 2000.
CROSS-SECTIONALSTUDY
COHORTSTUDY
1990 2000
41-50 41-50
51-60 '\.51-60
61-70 61-70
71-80 71-80
TRENDSTUDY
1990 2000
41-50 -41-50
51-60 51-60
61-70 -61-70
d
4ii 71-80 -71-80
PANELSTUDY
1990 2000
41-50" 41-50'
51-60" 51-60"
61-70X 61-70'
71-80' 71-80'
+%I'
-Denotescomparison
*Denotes same individuals
found that some studentshad tried both marijuana A doser examination of the data showed, how-
and LSD, some had tried only one, and others had ever, that although some students reported having
tried neither. Because these data were collected at tried marijuanabut not LSD,there were no stu-
one time, and because some studentspresumably dents in the studywho had tried onlyLSD. From
would experimentwith drugslater on, it would ap- thisfinding,the researchersinferred-as common
pear that such a study couldnot tell whether stu- sense suggested-that marijuana usepreceded LSD
dentswere more likelyto trymarijuana or LSD first. use. If the process of drug experimentationoccurred
Určeno pouze pro studijní účely
106 .Chapterrl: ResearchDesign HowTo Design a ResearchProject . 107
in the oppositetime order,then a study at a given
time shouldhave found some studentswho had
tried LSD but not marijuana, and it should have
fomd no studentswho had tried only marijuana.
Researcherscan also make logicalinferences
whenever the time order of variables is clear.If we
discover in a moss-sectional study of collegestu-
dents that those educatedin private high schools
receivedbetter college grades than did those edu-
cated in public high schools, we would conclude
that the type of high school attended affected col-
lege grades, not the other way around. Thus, even
though our observationswere made at only one
time, we would feeljustified in drawing conclu-
sions about processes that took place acrosstime.
Very often, age dierencesdiscovered in a
cross-sectionalstudy form the basis for inferring
processes aaoss time. Supposeyou are interested
in the pattern of worseninghealth over the course
of thc typical life cycle. You might pursue this sub-
ject by studying the results of annual checkupsin
a large hospital. You could group health records
according to the ages of those examined and rate
each age group in terms of severalhealth condi-
tions-sight, hearing, blood pressure, and so forth.
By reading aaoss the age-groupratings for each
health condition,you would have somethingap-
proximating the health history of individuals.Thus,
you might conclude that the averageperson devel-
ops vision problems before hearing problems. You
would need to be cautiousin this assumption, how-
ever, because the differencesmight reflect society-
wide trends. Perhaps improvedhearing exarnina-
tions institutedin the schools had affectedonly the
youngpeople in your study.
Asldng people to recall their pasts is another
common way of approximatingobservations over
time. Researchers use that method when they
ask people where they were born or when they
graduated from high school or whom they voted
for in 1996.Qualitativeresearchers often conduct
in-depth "lifehistory"interviews.For example,
C. Lynn Carr (1998)used this technique in a study
of "tomboyism."Her respondents,aged 25 to 40,
were asked to reconstructaspects of their lives from
childhood on, including experiences of identifyrng
themselves as tomboys.
The dangerin this teclmiqueis evident.Some-
times people have faultymemories; sometimesthey
lie.When people are askedinpostelectionpolls
whom they voted for, the results inevitably show
morepeople voting for the winner than actually
did so on election day. As part of a series of in-
depth interviews, such a report can be validated in
the context of other reported details;however, re-
sultsbased on a single questionin a survey must be
regarded with caution.
This discussionof the ways that timefigures
into social research suggest several questions you
should confrontin your own research projects.In
designingany study,be sure to examineboth the
explicitand the implicit assumptions you're making
about time. Are you interestedin describing or ex-
plaining some process that occurs over time, or are
you interestedsimply in what existsnow? TI you
want to describe a process occurring over time, will
you be able to make observations at differentpoints
in the process, or will you have to approximate
such observationsby drawinglogical inferences
from what you can observenow? If you opt for a
longitudinaldesign,which method best serves your
research purposes?
Examples ofResearchStrategies
As the preceding discussionshave implied, social
scientiiicresearchfollowsmany paths. The follow-
ing short excerptsfurther illustratethis point. AS
you read each excerpt,take note of both the con-
tent of each studyand the method used to study
the chosentopic. Does the study seem to be explor-
ing, desaibing, or explaining (orsome combination
of these)? What are the sourcesof datain each
study? Can you identify the unit of analysis? Is the
dimension of time relevant? If so, how willit be
handled?
This case study of unobtrusivemobilizing by
Southern California Rape Crisis Centeruses
archival, observational, and interview data to
explorehow a feminist organizationworked to
change police, schools,prosecutors, and some
stateand national organizationsfrom 1974to
1994.(SchmittandMartin 1999:364)
a Usinglife history narratives, the present study
investigatesprocesses of agencyand conscious-
ness among 14women who identifiedthem-
selvesas tomboys. (Carr1998:528)
e By drawing on interviews with activists in the
formerEstonian Soviet SocialistRepublic, we
specrfythe conditionsby which accornmoda-
tive and oppositionalsubculturesexist and are
successfullytransformedinto socialmove-
ments. (Johnstonand Snow 1998:473)
e Thispaperpresents the results of an ethno-
graphic study of an ADS senice organization
located in a smallcity. It is based on a combina-
tion of participant observation,interviewswith
participants, and review of organizational
records. (IGlburn1998:89)
s Usinginterviewsobtained duringfieldwork
in Palestinein 1992, 1993,and 1994,and
employinghistorical and archivalrecords, I ar-
gue that Palestinianfeminist discourseswere
shaped and influencedby the sociopolitical
context in which Palestinianwomen acted
and with which they interacted. (Abdulhadi
1998:649)
a, This articlereports on women's experiences
ofbreastfeedingin public as reveded through
in-depth interviews with 51women. (Stearns
1999:308)
s Usinginterview and observationalfield data,
I demonstratehow a system of temporary ern-
ployrnent in a participativeworkplace both
exploitedand shaped entry-levelworkers' as-
pirations and occupationalgoals. (Y Smith
1998:411)
a I collected data [on White SeparatistRhetoric]
from severalmedia ofpublic discourse, includ-
ing periodicals, books, pamphlets, transc~ipts
from radio and television talk shows, and
newspaper and magazine accounts. (Berbrier
1998:435)
e Inthe analysis that follows,racial and gender
inequality in employmentand retirement will
be analyzed,using a national sample of persons
who began receiving Social SecurityOld Age
benefits in 1980-81. (Hoganand Perrucci
1998:528)
e Drawingfrom interviews with female crack
dealers, this paper exploresthe techniques
they use to avoid arrest. (Jacobsand Miller
1998:550)
HowTo Design a Research Project
You've just seen some of the options availableto
socialresearchersin designingprojects. Now let's
pull the parts togetherby looking at the actual
process of designinga research project. Assume you
were to undertake a researchstudy.Where would
you start? Then, where would you go?
Although research design occurs at the begin-
ning of a research project, it involves allthe steps
of the subsequentproject. This discussion,then,
providesboth guidance on how to start a research
project and an overview of the topics that followin
later chapters of this book.
Figure 4-2 presents a schematicview of the
socialscienceresearch process. I present this view
;6luctantly, becauseit may suggest more of a step-
by-step order to research than actualpractice bears
out. Nonetheless, this idealized overviewof the
processprovides a context for the specificdetails
of particularcomponents of social research.Essen-
tially, it is another and more detailedpicture of the
scientiiicprocesspresented in Chapter 2.
At the top of the diagram are interest, idea, and
theory, the possiblebeginning pointsfor a line of
research. The letters (A,B, X, Y, and so forth) rep-
resent variables or concepts such as prejudice or
alienation.Thus, you might have a generalinterest
in kding out what causes somepeople to be more
prejudiced than others, or you might want to know
some of the consequencesof alienation. Alterna-
tively, your inquiry might begin with a specificidea
about the way things are. For example, you might
have the idea that workingin low-level servicejobs
(say,at a fast-food outlet) causes alienation.The
question marks in the diagramindicate that you
aren't sure things are the way you suspect they
are-that's why you're doing the research. Notice
that a theory is represented as a set of complex re-
lationships among severalvariables.
Určeno pouze pro studijní účely
114 .Chapter4: Research Design
you'll reach them. Will it be appropriate to select a
sample?If so, how will you do that? If there is any
possibility that your research will affectthose you
study, how will you insure that the research does
not harm them?
Measurement
What are the lceyvariables inyour study? How will
you d e h e and measure them? Do your demtions
and measurement methods duplicate or differfrom
those of previous research on this topic? Ifyou
have already developedyour measurement device
(a questiomaire, for example) or willbe using
sometlling previously developedby others, it might
be appropriateto include a copy in an appendix to
your proposal.
Data-Collection Methods
How willyou actually collect the data for your
study? Willyou conduct an experiment or a sur-
vey? W i you undertalcefield research or will you
focus on the reanalysis of statistics already created
by others? Perhapsyou will use more than one
method.
Analysis
Indicate the kind of analysisyou plan to conduct.
Spell out the purpose and logic of your analysis.
Are you interested in precise description? Do you
intend to explainwhy things are the way they are?
Do you plan to account for variations in some qual-
ity: for example,why some students are more lib-
eralthan others? What possible explanatoryvari-
ables will your analysis consider, and how willyou
know if you've explainedvariationsadequately?
Schedule
It is oftenappropriate to provide a schedulefor the
various stages of research. Even if you don't do this
for the proposal, do it foryourself. Unlessyou have
a timeline for accomplishingthe several stages of
research and keeping in touch with how you're do-
ing, you may end up in trouble.
Budget
When you ask someone to cover the costs of your
research, you need to provide a budget that speci-
fieswhere the money will go. Large, expensive
projects include budgetary categoriessuch asper-
sonnel, equipment, supplies,telephones, and
postage. Even for a project you pay for your-
self,it's a good idea to spend sometime anticipat-
ing expenses: office supplies,photocopying, com-
puter disks,telephone calls, fransportation, and
SO on.
As you can see,if you were interested in con-
ducting a social scienceresearchproject, it would
be a good idea to prepare a researchproposalfor
your own purposes, even if you weren't required
to do soby your instructor or a funding agency.If
you're going to invest your time and energy in such
a project,you should do what you can to insure a
return on that investment.
Now that you've had abroad overview of social
research, let's move on to the remaining chapters
in thisboolc and learn exactlyhow to design and
execute each specificstep.If you've found a re-
searchtopic that really interestsyou, you'll want to
keep it in mind as you see how you might go about
studyingit.
M A I N POINTS
e The principalpurposes of socialresearch in-
clude exploration,description, and explanation.
Research studies often combine more than one
purpose.
B) Exploration is the attempt to develop an initial,
rough understanding of somephenomenon.
e Descriptionis the precise reporting andlor
measurement of the characteristics of some
population or phenomenon under study.
Explanationis the discovery and reporting of
relationships among different aspects of the
phenomenon under study.Whereas descrip-
tive studies answer the question "What's so?"
explanatory ones tend to answer the ques-
tion "Why?"
Units of analysisare the people or things whose
characteristicssocialresearchers observe, de-
scribe, and explain. Typically,the unit of analy-
sisin socialresearchis the individualperson, but
it may also be a social group, a formal organi-
zation,a socialartifact, or some otherphenome-
non such as lifestyles or socialinteractions.
The ecologicalfallacyinvolves conclusions
drawn from the analysis of the attributes of
groups (e.g.,neighborhoods) that are then as-
sumedto apply to individuals (e.g.,specific res-
idents of different neighborhoods).
Reductionismis the attempt to understand
a complexphenomenon in terms of a narrow
set of concepts, such as attempting to explain
the American Revolution solely in terms of
economics (orpolitical idealismor psychology).
Researchinto processes that occur over time
presents social challenges that can be addressed
through cross-sectionalstudies or longitudinal
studies.
Cross-sectionalstudies are based on observa-
tions made at one time. Although such studies
are limited by this characteristic, researchers
can sometimesmake inferences about pro-
cesses that occur over time.
In longitudinal studies, observations are made
at many times. Such observationsmay be made
of samples drawn from generalpopulations
(trendstudies),samples drawn from more
specificsubpopulations (cohort studies),or
the same sample of people each time (panel
studies).
Research design starts with an initialinterest,
idea, or theoretical expectationand proceeds
through a series of interrelated steps to narrow
the focus of the study so that concepts, meth-
ods, and procedures are well defined.A good
research plan accountsfor all these steps in
advance.
At the outset, a researcher specifies the mean-
ing of the concepts or variablesto be studied
(conceptualization),chooses a research method
or methods (e.g.,experiments versus surveys),
and specifiesthe population to be studiedand,
if applicable, how it will be sampled.
Review Questions and Exercises .115
s The researcher operationalizesthe concepts to
be studied by stating precisely how variables in
the study will be measured. Researchthen pro-
ceeds through observation,processingthe data,
analysis, and application,such as reporting the
results and assessingtheir implications.
e A research proposal provides a preview of why
a study willbe undertalcen and how it will be
conducted.A researchproject is often required
to get permission or necessaryresources.Even
when not required, a proposalis a useful de-
vice for planning.
KEY TERMS
units of analysis longitudinal study
socialartifact trend study
ecologicalfallacy cohort study
reductionism panel study
cross-sectionalstudy
L: REVIEW QUESTIONS AND EXERCISES
1. Using InfoTrac or the library, select a research re-
port that illustratesexploration,description,or ex-
planation. Identifywhich of these three purposes
the report illustratesand brieflyjustify yourjudg-
ment in that regard.
2. Here are some examples of real research topics.
For each one, name the unit of analysis. (Thean-
swersare at the end of this chapter.)
a. Women watch TV more than men because
they are likelyto worlcfewer hours outside
the home than men. ...Black people watch
an average of approximatelythree-quartersof
an hour more television per day than white
people. (Hughes1980:290)
b. Of the 130incorporatedU.S. cities with more
than 100,000inhabitantsin 1960, 126had at
least two short-termnonproprietarygeneral
hospitals accreditedby the American Hospital
Association. (Turk 1980:317)
c. The earlyTM [transcendentalmeditation]
organizationswere small and informal.The
LosAngeles group,begun in June 1959,met
at a member's house where, incidentally,
Maharishi was living. (Johnston1980:337)
Určeno pouze pro studijní účely
116 .Chapter4: Research Design
d. However, it appears that the nursing staffs
exercisestronginfluence over. ..a decision
to change the nursing care system. ...Con-
versely, among those decisionsdominatedby
the administrationand the medical staffs ...
(Comstock 1980:77)
e. Though 667,000 out of 2 millon farmersin
the United Statesare women, women histori-
cally have not been viewed as farmers,but
rather, as the farmer's wife. (Votaw 1979:8)
f. The analysisof community opposition
to group homes for the mentallyhandi-
capped ...indicatesthat deteriorating neigh-
borhoods are most likelyto organizein
opposition,but that upper-middleclass neigh-
bbrhoods are most likely to enjoyprivate ac-
cess to local officials. (Grahamand Hogan
1990:513)
g. Some analysts during the 1960spredicted
that the rise of economicambitionand po-
liticalmilitancy among blacks would foster
discontentwith the "otherworldly"black
mainline churches. (Ellisonand Sherkat
1990:551)
h. This analysisexploreswhether propositions
and empirical mdings of contemporarytheo-
ries of organizationsdirectlyapplyto both pri-
vate prod~~ctproducing organizations (PPOs)
and publichuman serviceorganizations
(PSOs).(Schitlettand Zey 1990:569)
i. Thispaper examinesvariationsinjob title
structuresacross work roles. Analyzing
3,173job titles in the California civil service
systemin 1985,we investigatehow and why
lines of work vary in the proliferation of job
categoriesthat differentiate ranks, functions,
or particular organizationallocations. (Strang
and Baron 1990:479)
3. Look through a n academic researchjournal until
you find examples of at leastthree differentunits
of analysis. 1den.Qeach and present quotations
from thejournal to justify your conclusions.
4. Make up a research example-different from
those discussed in the text-that illustratesa re-
searcher faUing into the trap of the ecologicalfal-
lacy. Then modify the exampleto avoid this trap.
5. Drop in at the Russell SageFoundation
(m)and look at their
publications. Select one that illustratesa cross-
sectional,trend, cohort, or panel study design.
Justify your choice.
ADDITIONAL READINGS
Bart, Pauline, and Linda Frankel. 1986. TheStudent
Sociologist'sHandbook. Morristown, NJ: General
LearningPress.A handy little reference book to
help you get started on a researchproject. Written
from the standpoint of a student term paper, this
volume offers a particularly good guide to the pe-
riodical literatureof the social sciencesavailable in
a good library.
Casley,D. J., andD. A. Lury.1987.Data Collectionin
Deve1opi)zg Coulitries. Oxford: ClarendonPress. This
boolc discussesthe specialproblems of researchin
the developingworld.
Cooper,HarrisM. 1989.Integrating Research:A Guide
for Literature Reviews. Newbury Park, CA: Sage.The
author leads you through each stepin the litera-
ture reviewprocess.
Hunt, Morton. 1985.Profilesof SocialResearch: The Sci-
entifcStudy ofHuman Interactions. New York: Basic
Books. An engagingand informative seriesof
project biographies: James Coleman's smdyof
segregated schoolsis presented, as well as several
other major projects that illustrate the elements
of socialresearch in practice.
Iversen, GudmundR. 1991.Contextual Analysis. New-
bury Park, CA: Sage. Contextual analysis exarn-
ines the impact of socioenvironmentalfactors
on individualbehavior. Durkheim's study of sui-
cide offers a good example of this, identifying
socialcontexts that affect the likelihood of self-
destruction.
Maxwell, Joseph A. 1996.Qtlalitative Research Deshn:
An Interactive Approaclz. Newbury Park, CA: Sage.
Maxwell covers many of the sametopicsthat this
chapter doesbut with attention devoted specifi-
cally to qualitativeresearch projects.
Menard, Scott. 1991.Longitudinal Research. Newbury
Park, CA: Sage. Beginningby explainingwhy re-
searchersconductlongitudinal research,the au-
thor goes on to detail a variety of study designs as
well as suggestions for the analysis of longitudinal
data.
Miller, Delbert. 1991.Handbook of Research Design and
Social Measurement. Newbury Park, CA: Sage. A
usefulreference for introducingor reviewing nu-
merousissues involved in design and measure-
ment. In addition, the book containsa wealth of
practical informationrelating to foundations,
journals, andprofessionalassociations.
ANSWERS TO REVIEW QUEST10NS
AND EXERCISES, ITEM 2
a. Men and women, black and white people
(individuals)
b. Incorporated U.S. cities (groups)
C. Transcendentalmeditation organizations (groups)
d. Nursing staffs (groups)
e. Farmers (individuals)
f. Neighborhoods (groups)
g. Blacks (individuals)
h. Serviceand productionorganizations (formal
organizations)
i. Job titles (artifacts)
SOCIOLOGY WEB SITE
See the Wadsworth SociologyResource Center,
4 8 a A .s ~ . vVutual Society, for additionallinks,hternet ex-
ercisesby chapter, quizzes by chapter, and Microcase-
related materials:
Answers to Review Questionsand Exercises, Item 2 .117
lNFOTRAC COLLEGE EDITION
SEARCH WORD SUMMARY
Go to the Wadsworth SociologyResource Cen-
@ ter, Virtual Society,to find a list of searchwords
for each chapter. Using the search words, go to Wo-
Trac College Edition, an online library of over 900
journals where you can do onlineresearch and find
readings related to your studies. To aid in your search
and to gain useful tips, see the Student Guide to Info-
Trac College Edition on the Virtual SocietyWeb site:
Určeno pouze pro studijní účely