Analyses of Social Issues and Public Policy, Vol. 12, No. 1, 2.docx
1. Analyses of Social Issues and Public Policy, Vol. 12, No. 1,
2012, pp. 296--311
Theories of Radicalization
Testing Theories of Radicalization in Polls of U.S.
Muslims
Clark McCauley∗
Bryn Mawr College
Four national polls of Muslim Americans conducted between
2001 and 2007 were
reviewed to find items tapping possible sources of sympathy
and justification for
jihadist violence: anti-Muslim discrimination, radical Islam, and
economic and
political grievance. These items were correlated with items
representing three
elements of the global-jihad frame: seeing the war on terrorism
as a war on Islam
or “insincere,” justifying suicide attacks in defense of Islam,
and favorable views
of Al Qaeda. The three elements of the global-jihad frame were
no more than
weakly related to one another and had different predictors.
Discussion suggests
that the U.S. “war of ideas” may need to target separately the
different elements
of the global-jihad frame.
Attempts to understand terrorism usually begin with the
2. terrorists, and a
sizeable literature has explored whether or how terrorists
emerge from conditions
of poverty, alienation, extremist ideology, and political
grievance (Blair, 2009;
Brown & Smith, 2009; Krueger, 2007; Sageman, 2004). But
active terrorists are
only a tiny fraction of the base of supporters and sympathizers
on which the
terrorists depend, just as active-service military are only a tiny
fraction of the base
of national support for the military. In this article, I follow
Tessler and Robbins
(2007) in moving the focus away from terrorists to the wider
population of those
who sympathize with or justify terrorism.
∗ Correspondence concerning this article should be addressed to
Clark McCauley, Psychology
Department, Bryn Mawr College, Bryn Mawr, PA 19010 [e-
mail: [email protected]].
The author thanks Dennis Gilbert for generously sharing the
data of the Hamilton College Muslim
American Poll, and thanks Gary LaFree and Susan Brandon for
suggestions toward improving the
research. This research was supported by the U.S. Department
of Homeland Security through the
National Consortium for the Study of Terrorism and Responses
to Terrorism (START), grant number
N00140510629. However, any opinions, findings, conclusions,
or recommendations in this report are
those of the author and do not necessarily reflect views of the
U.S. Department of Homeland Security.
296
4. with elements of the global-jihad frame from respondents who
do not agree.
Public Opinion and the Global-Jihad Frame
It is important to recognize that radicalization of opinion is not
the same as
radicalization of behavior. A poll of U.K. Muslims after the
July 7, 2005, suicide
bombings in the London Underground asked, “Do you think any
further attacks
by British suicide bombers in the U.K. are justified or
unjustified?” Five percent
of poll respondents thought that further bombings would be
justified, a result that
implies that about 50,000 of 1 million adult U.K. Muslims
believed further bomb-
ings would be justified. But, between 2001 and 2010, terrorism-
related charges in
the United Kingdom have been brought against only 383
individuals (BBC News,
2010). Even if we assume that U.K. security has missed one
terrorist for every
one apprehended, these data indicate that, of every hundred
U.K. Muslims con-
doning suicide attacks, at most two are actively involved in
terrorism. The great
majority of those condoning attacks are doing nothing (see
Leuprecht, Hataley,
Moskalenko, & McCauley, 2010, for more about the distinction
between opinion
and action).
Similarly, results of the polls studied by Tessler and Robbins
(2007) indicate
the importance of distinguishing between terrorists and their
5. sympathizers. Most
jihadist terrorists are male, for instance, but polls in Algeria and
Jordan found males
no more likely than females to approve of terrorism. In
addition, active terrorists
can be above average in education and income in relation to the
communities they
emerge from—as Krueger (2007) found for Palestinian and
Jewish terrorists—
even as they depend on a base of sympathizers and supporters
less advantaged
than themselves.
A pyramid model linking a small apex of terrorists with a larger
base of sym-
pathizers and supporters (Moskalenko & McCauley, 2009) is
consistent with the
298 McCauley
U.K. government’s definition of radicalization: “Radicalisation
. . . is the process
by which people come to support terrorism and violent
extremism and, in some
cases [emphasis added], then to participate in terrorist groups”
(Brown & Smith,
2009, p. 43).
There is evidence indicating that public opinion can have an
effect on terrorism
and terrorists. Krueger (2007) has used polling data to show that
terrorists are likely
to come from countries where higher percentages disapprove of
the leadership of
6. the country terrorists attack. Terrorists attacking the United
States, for instance,
are likely to come from countries with higher percentages who
disapprove of the
job performance of the leadership of the United States
A different kind of evidence for the effect of public opinion
comes from two
case studies of “sudden desistence” from terrorism: the
Armenian Secret Army for
the Liberation of Armenia, and the Egyptian Islamic Group. For
both cases, there
was reason to conclude that fast decline in terrorist attacks
could be attributed, at
least in part, to sudden loss of sympathy and support for
terrorism in the broader
public on whom terrorists depended (Dugan, Huang, LaFree, &
McCauley, 2009;
Wheatley & McCauley, 2009).
If mass radicalization means increasing numbers who
sympathize with terror-
ist goals and justify terrorist means, it remains to ask what
specific opinions signal
radicalization in relation to jihadist terrorism. In simplest its
form, the jihadist
frame is that the West is engaged in a millennial battle against
Islam and Muslims
must defend themselves—Islam is under attack and Muslims
have an obligation to
rise to its defense (for substantiation from primary Taliban
sources, see Johnson,
2007). Betz (2008, p. 520) offers a more detailed version of the
global-jihad frame:
(1) Islam is under attack by Western crusaders led by the United
States; (2) Jihadis,
7. called “terrorists” in the West, are defending against this attack;
(3) the actions
they take in defense of Islam are proportionally just and
religiously sanctified;
and, therefore (4) it is the duty of good Muslims to support
these actions.
The elements of the global-jihad frame are usually found
together in the
beliefs of jihadist terrorists, including many of the 172 U.S.
Muslims indicted in
43 terrorist plots identified since 9/11 (Bejelopera & Randall,
2010; Bergen &
Hoffman, 2010). It is not clear whether these beliefs are found
together in the
larger population of 1 million U.S. Muslims. An important
question raised in the
research reported here is the extent to which these beliefs co-
occur in polls of U.S.
Muslims.
The present study depends on data from polls already
conducted. In these
polls, agreement with the global-jihad frame is assessed by
focusing on three
kinds of items: items assessing perception that the war on
terrorism is a war on
Islam or “insincere” (Betz 1), an item assessing attitude toward
Al Qaeda (Betz
2), and an item assessing justification for suicide terrorism in
defense of Islam
(Betz 3). It seems that no poll has asked U.S. Muslims if they
felt a personal
responsibility to support jihadist violence (Betz 4).
8. Testing Theories of Radicalization in Polls of U.S. Muslims 299
The next three sections turn to possible predictors of accepting
the three
elements of the global-jihad frame for which I found items in
published polls.
Although these predictors could be derived from psychological
theories (Tessler
& Robbins, 2007), I focus here on ideas prominent in analyses
published by the
U.S. and U.K. governments.
Perceived Discrimination and Alienation
Two prominent intellectuals have advanced the idea that Muslim
radicaliza-
tion, especially in Europe, is a product of political alienation.
Olivier Roy (2004)
sees radical Islam, the kind that supports jihadi terrorists, as
emerging from modern
challenges to traditional Muslim identity. He sees these
challenges as especially
acute for Muslims who emigrate to Western countries. Francis
Fukuyama (2006)
puts a twist on the same idea to explain why Europe’s Muslims
are more alienated
and, he believes, more prone to terrorism than U.S. Muslims. In
brief, Fukuyama’s
argument is that Muslims in Western countries, especially in
Europe, adopt Osama
bin Laden’s global umma as an identity to the extent that ethnic
nationalism blocks
their acceptance in the country they live in. Identification with
the umma brings sus-
ceptibility to bin Laden’s claim that Islam is under global attack
9. by Western forces,
and the result of this frame is the justification of violence in
defense of Islam.
Discrimination and alienation are also featured in the analysis
of terrorism
developed in 2009 by U.K. security agencies.
In many countries, including the UK, people are not only
vulnerable to radicalisation
because of political and economic grievances. A range of social
and psychological factors
are also important. Radicalisation seems to be related directly to
a crisis in identity and,
specifically, to a feeling of not being accepted or not belonging.
This is itself the result of
a range of factors, which may include the experience of
discrimination and inequalities,
racism, recent migration and more generally a lack of affinity
with and disconnect from
family, community and state (Brown and Smith, 2009).
Another version of the alienation argument is that Muslims in
Western coun-
tries are rejecting, or are being encouraged to reject, the
integration that is in
fact offered them. This version can be found in the U.S. 2009
Annual Threat
Assessment of the Intelligence Community for the Senate Select
Committee on
Intelligence.
The social, political, and economic integration of Western
Europe’s 15 to 20 million Mus-
lims is progressing slowly, creating opportunities for extremist
propagandists and recruiters.
10. The highly diverse Muslim population in Europe already faces
much higher poverty and
unemployment rates than the general population, and the current
economic crisis almost
certainly will disproportionately affect the region’s Muslims.
Numerous worldwide and
European Islamic groups are actively encouraging Muslims in
Europe to reject assimilation
and support militant versions of Islam (Blair, 2009, p. 5).
Thus, two versions of alienation theory can be distinguished:
that Muslims
in Western countries see themselves as a minority discriminated
against by the
300 McCauley
majority, and that Muslims in Western countries are a minority
determined not to
melt away into the majority. Either way, Muslims in Western
countries are likely
to see themselves as suffering from bias and discrimination:
either because the
majority does not accept them or because they interpret pressure
to integrate as
bias and discrimination. In the polls examined here, items
tapping perception of
bias and discrimination against Muslims are included as
potential predictors of
agreement with the global-jihad frame.
Radical Islam
A second major idea about Muslim radicalization is that an
11. extremist form
of Islam is the major driver of Muslim radicalization and
jihadist terrorism. This
idea has been particularly popular in the U.S. military (but see
Smart, 2005, who
acknowledges this popularity even as she argues that ideology is
more justification
than provocation of violence). Indeed radical Islam is often
described in military
terms as the terrorist “center of gravity”—the primary source of
enemy power
(Eikmeier 2007, p. 86). If the ideological center of gravity can
be destroyed, the
enemy is finished (Lloyd, 2003).
The popularity of an ideological view of Muslim radicalization
has probably
increased thanks to an unusual film, Obsession. Produced in a
documentary style,
the film highlights parallels between the power of political
Islam and the power
of Nazi ideology. The film has been widely distributed,
including millions of
DVDs given away as inserts in major U.S. newspapers before
the 2008 election
(JewsOnFirst, 2008).
Radical Islam goes to the roots of life, prescribing not just
politics but personal
and public morality, spiritual life, and religious expression in a
life-dominating
“Puritan Islam.” Individuals participating in the ideology of
radical Islam should
show high religiosity, including reporting religion as an
important part of their
life and frequent participation in prayers at their mosque or
12. Islamic center. In the
polls examined here, these religiosity items are included as
potential predictors of
agreement with the global-jihad frame.
Economic and Political Grievance
Perhaps the most intuitive idea about political radicalization is
that it is a
response to some kind of grievance. This idea is represented in
the U.S. State
Department’s Country Reports on Terrorism 2008: “Rather, we
saw increasing
evidence of terrorists and extremists manipulating the
grievances of alienated
youth or immigrant populations, and then cynically exploiting
those grievances
to subvert legitimate authority and create unrest.” Similarly,
according to the
Office for the Coordinator for Counterterrorism (2009), “Efforts
to manipulate
grievances represent a ‘conveyor belt’ through which terrorists
seek to convert
Testing Theories of Radicalization in Polls of U.S. Muslims 301
alienated or aggrieved populations, by stages, to increasingly
radicalized and
extremist viewpoints, turning them into sympathizers,
supporters, and ultimately,
in some cases, members of terrorist networks.”
What are the grievances that can move Muslims toward
radicalization? A com-
13. mon view is that radicalization and terrorism arise out of
economic frustration. In
the polls examined here, this idea is tested by using standard
demographic mea-
sures of education and family income to predict opinions
representing elements of
the global-jihad frame. Because most terrorists are young and
male, I also look at
age and gender as potential predictors of agreement with the
global-jihad frame.
Another kind of grievance is political: U.S. policies in relation
to the Islamic
world may be a source of radicalization for many Muslims
(Tessler & Robbins,
2007). According to Scheuer (2004, 2006), the major grievances
behind jihadist
terrorism are U.S. troops in Muslim countries and U.S. support
for authoritarian
governments in Muslim countries. In the polls examined here,
this idea is tested
by using an item about opposition to U.S. forces in Afghanistan
as a potential
predictor of agreement with elements of the global-jihad frame.
In sum, the current study tests three broad ideas about the
sources of Mus-
lim radicalization—anti-Muslim discrimination, radical Islam,
and economic or
political grievances—in polls of U.S. Muslims that included
measures of agree-
ment with three elements of the global-jihad frame: seeing the
war on terror-
ism as a war on Islam, support for suicide bombing, and
favorable opinion of
Al Qaeda.
14. Methods
The data analyzed come from four national polls of U.S.
Muslims conducted
between 2001 and 2007. Zogby 2001 and Zogby 2004 polls were
purchased;
Zogby 2002 was made available by Dennis Gilbert at Hamilton
College, and Pew
2007 is available by download (http://people-
press.org/dataarchive/). Only brief
descriptions of these polls are provided here, additional details,
including sampling
methods, are available from Zogby and Pew.
Four Polls of U.S. Muslims (2001–2007)
Zogby 2001: The American Muslim Poll (n = 1,781) was
conducted in
November and December 2001 by the Muslims in American
Public Square
(MAPS) supported by the Pew Research Center
(http://pewresearch.org/about/)
in collaboration with Zogby International
(http://www.zogby.com/). In this and
other polls considered here, all respondents were 18 years of
age or older. A
telephone list was created by matching the Zip codes for 300
Islamic centers na-
tionwide against their respective local telephone exchanges;
listings of common
Muslim surnames were then identified from the local telephone
exchanges and
15. 302 McCauley
a random sample of names were called using random digit
dialing (RDD). An
additional sample of African-American Muslims was obtained
in face-to-face in-
terviews conducted December 7–9, 2001, at locations in New
York, Washington,
DC, Atlanta, GA, and Detroit, MI.
Zogby 2002: The Hamilton College Muslim American Poll (n =
521) was
designed by Sociology Professor Dennis Gilbert and a team of
Hamilton students
and supported by the Arthur Levitt Public Affairs Center. It was
conducted in April
2002 in collaboration with Zogby International. A national call
list was created by
software that identifies common Muslim names in telephone
listings.
Zogby 2004: The American Muslim Poll (n = 1,846) was
conducted in
August and September 2004 by MAPS supported by the Pew
Charitable Trusts in
conjunction with Zogby International. Telephone interviews
were carried out with
a nationwide sample of American Muslims using the same
methods as for Zogby
2001, except that no additional face-to-face sample was
included.
Pew 2007: The Muslim Americans Survey (n = 1,050) was
conducted by
Schulman, Ronca, and Bucuvalas Incorporated (SRBI) between
January and April
16. of 2007, according to the specifications of the Pew Research
Center. The sampling
frame had three parts: two RDD samples and one recontact
sample of Muslims
identified in earlier polls. The complexities and advantages of
this multistrata
polling are detailed in pages 57–71 of the poll report (Pew
Research Center, 2007).
Variation in Item Wording Across Polls
For some items, there are slight variations in wording of
question or response
alternatives across polls. These variations probably do not
affect the substantive
meaning of the item, but the variations are presented so that
readers can judge for
themselves. In any case, the focus here is on correlations among
items rather than
on change over time in responses to the same or similar items,
and consistent cor-
relations across polls despite variation in item wording indicate
that the variations
have little impact on results.
Results
The first section of the Results presents the distribution of
opinion for three
criterion items, the second section presents the distribution of
opinion for five
predictor items, and the third and fourth sections present
correlational analyses
linking predictors with criterion items.
Criterion Items
17. Table 1 shows that substantial percentages of U.S. Muslims see
the war on
terrorism as a war on Islam or don’t believe the war is a sincere
effort to reduce
Testing Theories of Radicalization in Polls of U.S. Muslims 303
Table 1. Responses of U.S. Muslims to Items Assessing Attitude
Toward the U.S. War on Terrorism
Zogby 2001a Zogby 2002b Zogby 2004c Pew 2007d
(N = 1,781) (N = 531) (N = 1,846) (N = 1,050)
War on terrorism (sincere effort) 67 41 33 26
War on Islam (not sincere effort) 18 31 38 55
Neither/both – 20 – 2
NS/DK/R 16 7 29 17
Note: Pew item is less extreme than Zogby items; Pew results
not directly comparable to Zogby
results.
a In the aftermath of the September 11 attacks, do you feel the
United States is fighting a war on
terrorism or a war against Islam?
b Some describe the U.S. worldwide response to the September
11 attacks as a war on terrorism.
Others say it is a war on Islam. Which do you think is more
accurate?
c Do you feel the United States is fighting a war on terrorism or
a war against Islam?
d Do you think the U.S.-led war on terrorism is a sincere effort
to reduce international terrorism or
don’t you believe that?
18. terrorism. Believing that the war on terrorism is a war on Islam
(Zogby items) is
a more extreme version of disapproval than believing the war is
not sincere (Pew
item), and it is not easy to interpret the difference between
Zogby 2004 (38% see
war on Islam) and Pew 2007 (55% do not believe war on
terrorism is sincere).
Nevertheless, both the Zogby items and the Pew item tap
disapproval of the war
on terrorism, and correlates of disapproval will be of interest
for both Zogby and
Pew items.
Two additional criterion items are available only in the 2007
Pew poll, one
about suicide bombing and one about Al Qaeda.
Some people think that suicide bombing and other forms of
violence against
civilian targets are justified in order to defend Islam from its
enemies. Other people
believe that, no matter what the reason, this kind of violence is
never justified.
Do you personally feel that this kind of violence is often
justified to defend Islam,
sometimes justified, rarely justified, or never justified?
Responses were as follows:
1 = never (82%), 2 = rarely (4%), 3 = sometimes (6%), 4 =
often (2%), and
not sure/don’t know/refused (NS/DK/R) (6%). To ensure
comparable correlations
based on the same respondents, NS/DK/R were recoded as 2.5.
The second criterion item, available only in Pew 2007, was:
19. Overall, do you
have a favorable or unfavorable opinion of Al Qaeda? 1 = very
unfavorable
(67%), 2 = somewhat unfavorable (8%), 3 = somewhat favorable
(3%), 4 = very
favorable (1%), NS/DK/R (22%). Here again, NS/DK/R were
recoded as 2.5. For
both Pew-only items, correlations with NS/DK/R recoded (Table
3) never differed
by more than .03 from the corresponding correlations (not
tabled) with NS/DK/R
treated as missing data.
The two Pew-only items have skewed distributions, with the
great majority
of respondents disapproving of suicide bombing (86%) and
unfavorable toward
Al Qaeda (75%). In addition, the Al Qaeda item has a
substantial percentage
304 McCauley
(22%) not answering the question and recoded as neutral
(neither favorable nor
unfavorable). These item characteristics imply that the items
will be difficult
to predict, but the significant correlation between the two Pew-
only items is
encouraging. Respondents favorable toward Al Qaeda are more
likely to justify
suicide bombing, r = .30 with recodes (r = .32 without recodes).
It is interesting
to note that opinions about suicide bombing and about Al Qaeda
show negligible
20. correlations (.04 and .08 with recodes, .14 and .08 without
recodes) with opinion
about whether the war on terrorism is “sincere.”
Predictor Items
Standard demographic items included in all four polls included
gender, age,
education, family income, and birthplace (United States or
other).
Religiosity. Zogby: Would you say the role of Islam in your life
is very
important, somewhat important, or not very important? Pew:
How important is
religion in your life?
All four polls show that the importance of Islam in the lives of
U.S. Muslims
is consistently high: 79%, 81%, 82%, and 72% very important
(NS/DK/R = 1%).
Zogby 2001 and 2004: On average, how often do you attend the
mosque
for salah and Jum’ah Prayer? Zogby 2002: Approximately how
often do you
attend a mosque for prayer? Pew 2007: On average, how often
do you attend the
mosque or Islamic center for salah and Jum’ah prayer? About
half of Muslim
respondents report attending mosque at least once a week: 55%,
51%, 54%, 40%
(NS/DK/R = 1%).
Perceived discrimination. Zogby: Do you think the media is fair
in its por-
21. trayal of Muslims and Islam? Pew: Do you think that coverage
of Islam and
Muslims by American news organizations is generally fair or
unfair? Over half
of U.S. Muslims believe that Islam and Muslims are unfairly
portrayed in the me-
dia: Zogby 2001 (77%), Zogby 2004 (76%), and Pew 2007
(57%) (NS/DK/R =
7–10%). There was no similar question in Zogby 2002.
Zogby 2002: Aside from restrictions on religious expression at
work, have you
yourself suffered anti-Muslim discrimination, harassment,
verbal abuse, or phys-
ical attack since September 11? Zogby 2004: Have you yourself
experienced any
discrimination of this sort since 9/11? Pew: And thinking more
generally—NOT
just about the past 12 months—have you ever been the victim of
discrimination as
a Muslim living in the United States? At least a quarter of U.S.
Muslims report ex-
perience of personal discrimination: Zogby 2002 (26%), Zogby
2004 (40%), Pew
2007 (27%) (NS/DK/R = 1%). There was no similar question in
Zogby 2001.
Political grievance. Zogby 2001 and 2004: Do you strongly
support, some-
what support, somewhat oppose, or strongly oppose the U.S.
military action
Testing Theories of Radicalization in Polls of U.S. Muslims 305
22. Table 2. Correlations (Regression Betas) with Seeing the War
on Terrorism as a War on Islam
(Zogby Polls)
Zogby 2001 Zogby 2002 Zogby 2004
(N = 1,781) (N = 521) (N = 1,846)
Demographics
Gender (male = 1, female = 2) .03 (–.01) .13 (.07) .10 (.05)
Age −.10 (.00) .02 (.06) −.03 (.04)
Education −.01(−.05) .05 (.02) −.04 (−.05)
Family income −.03 (−.03) .04 (.05) −.02 (.04)
Born in the United States (no = 1, yes = 2) .18 (.07) .08 (.03)
.18 (.07)∗
Religiosity
Islam in your life (not important = 1, very
important = 3)
.14 (.01) .16 (−.06) .15 (.07)∗
Attend mosque (never = 1, more than
once/wk = 6)
.16 (.05) .12 (.07) .11 (.01)
Discrimination
American media fair in portrayal of
Muslims and Islam (yes = 1, no = 2)
.23 (.11)∗ – .27 (.16)∗
Personally suffered anti-Muslim
discrimination since September 11
(no = 1, yes = 2)
23. – .22 (.17)∗ .18 (.11)∗
Grievance
U.S. military action in Afghanistan
(strongly support = 1, strongly oppose =
4, other = 2.5)
.52 (.46)∗ .46 (.41)∗ .40 (.32)∗
∗ Beta different from zero, p < .01. For 2001, R(1.282) = .52;
for 2002, R(508) = .51; for 2004,
R(1,527) = .47
Note: Tabled correlations with war on terrorism or war against
Islam (terrorism = 1, Islam = 2,
other = 1.5). Owing to missing responses, n of 2001 correlations
ranged from 1,655 to 1,781 (except
income n = 1,480); n of 2002 correlations 513–521; n of 2004
correlations 1,790–1,846 (except
income n = 1,591). See Table 1 for variations in wording of the
criterion item; see text for variations
of some predictor items.
against Afghanistan? Zogby 2002: U.S. military action in
Afghanistan after
September 11 was justified under the circumstances. Do you
strongly agree, some-
what agree, somewhat disagree, or strongly disagree? Pew 2007:
Do you think
the United States made the right decision or the wrong decision
in using military
force in Afghanistan? Between one half and one third of U.S.
Muslims opposed
U.S. military action in Afghanistan: 52%, 53%, 35%, 35%
(NS/DK/R = 6–17%).
24. Predicting Opinions About the War on Terrorism
Table 2 shows correlations of demographic and opinion items
with seeing
the war on terrorism as a war on Islam (Zogby items), and Table
3 shows
the same predictors correlated with seeing the war on terrorism
as “insincere”
(Pew item).
306 McCauley
Table 3. Correlations (Regression Betas) with Seeing War on
Terrorism Insincere, with Justifying
Suicide Bombing, and with Sympathy for Al Qaeda (Pew 2007
Poll)
War on Suicide bombing Favorable
terrorism justified to opinion of
not sincere defend Islam Al Qaeda
Demographics
Gender (male = 1, female = 2) .08 (.00) .05 (−.01) .10 (−.02)
Age −.04 (−.05) −.14 (−.13)∗ −.05 (−.02)
Education .00 (−.01) −.16 (−.13)∗ −.29 (−.17)∗
Family income .06 (.07) −.09 (.02) −.22 (−.06)
Born in the United States (no = 1, yes = 2) .19 (.12)∗ .03 (.03)
.11 (.07)
Religiosity
Religion in your life (not important = 1, very
important = 3)
25. .08 (.02) .05 (.02) .13 (.03)
Attend mosque (never = 1, more than
once/wk = 6)
.06 (−.02) .04 (.02) .09 (.07)
Discrimination
Coverage of Islam and Muslims by American
news organizations (fair = 1, unfair = 2,
depends = 1.5)
.25 (.17)∗ −.05 (−.03) −.08 (−.08)
Ever victim of discrimination as a Muslim
living in the United States (no = 1, yes = 2)
.17 (.08) .04 (.06) .02 (.00)
Grievance
Military force in Afghanistan (right = 1,
wrong = 2, other = 1.5)
.31 (.25)∗ .08 (.08) .22 (.19)∗
∗ Beta different from zero, p < .01. For war on terrorism,
R(798) = .39; for suicide bombing,
R(798) = .22; for favoring AQ, R(798) = .37.
Note: First column: sincere effort = 1, don’t believe that = 2,
other = 1.5. Second column: never =
1, often = 4, other = 2.5. Third column: very unfavorable = 1,
very favorable = 4, other = 2.5. N
of tabled correlations 1,024–1,050, except n = 959 for media
item and n = 868 for income item. See
text for complete wording of predictor and criterion items.
26. The first thing to notice is the consistency of the pattern of
correlations across
the four surveys (three columns of Table 2, first column of
Table 3). Demographic
predictors—gender, age, education, family income—are
consistently uncorrelated
with seeing the war on terrorism as a war on Islam. Being born
in the United States
is weakly but consistently associated with seeing a war on Islam
(magnitude of
correlations ranging from .08 to .19). The two religiosity
items—importance of
Islam and frequency of mosque attendance—show consistent
correlations, with
higher religiosity associated with more disapproval of the war
on terrorism. These
correlations are weak, however (magnitude ranging from .06 to
.16). The two
perceived discrimination items also show consistent
correlations, with more per-
ceived discrimination associated with more disapproval of the
war on terrorism.
These correlations are a little stronger than the religiosity
correlations but still
weak (magnitude ranging from .17 to .25). Finally, disapproval
of the war in
Afghanistan shows consistently the strongest correlations with
disapproval of the
war on terrorism (magnitude ranging from .31 to .52).
Testing Theories of Radicalization in Polls of U.S. Muslims 307
In order to examine further the predictability of opinions about
27. the war on
terrorism, I conducted multiple regression (R) for each of the
four polls, entering
all tabled predictors in one step. Multiple Rs were .52 in 2001,
.51 in 2002, .47 in
2004, and .39 in 2007. R betas that show the unique
contribution of each predictor
appear in parentheses in Tables 2 and 3.
The betas are very like the corresponding zero-order
correlations, and simi-
larly consistent across polls. Demographic and religiosity items
are useless or weak
predictors, although betas show a small but consistent tendency
for respondents
born in the United States to be more likely to disapprove of the
war on terrorism.
Perception of personal or group discrimination provides weak
prediction of dis-
approval of the war on terrorism (beta magnitudes .11 to .17).
Only opposition to
the U.S. military in Afghanistan shows consistently strong
prediction of seeing a
war on Islam (betas .46, .41, .32, and .25).
Predicting Opinions About Suicide Bombing and Al Qaeda
Table 3 shows that the best (but weak) predictors of seeing
suicide bombing
as justified are two demographic items, age and education (–.14,
–.16). These
correlations indicate that there is a small but consistent
tendency for younger and
less-educated respondents to justify suicide bombing more than
older and more
educated respondents. R betas (–.13, –.13) confirm that only age
28. and education
are significant predictors of justifying suicide bombing,
although the multiple R
is only .22.
The best correlates of favorable opinion of Al Qaeda are
education (–.29),
family income (–.22), and opposition to the war in Afghanistan
(.22). These
correlations indicate that respondents with lower education and
lower income are
more likely to have favorable opinions of Al Qaeda, and that
respondents more
opposed to the Afghan war are more likely to have favorable
opinions of Al Qaeda.
R betas (–.17, .19) indicate that only low education and
opposition to U.S. forces
in Afghanistan are reliable predictors of favoring Al Qaeda. The
multiple R is .37.
Thus, more positive opinions about suicide bombing and about
Al Qaeda are
both predicted by lower education, but only opinions about Al
Qaeda are predicted
by opposition to the Afghan war. Given that the correlation
between the suicide
bombing item and the Al Qaeda item is .30, it is not surprising
that these two items
can have different predictors.
Discussion
Political radicalization of Muslims in Western countries has
generally been
given three kinds of explanation. The first explanation is
alienation: the idea
29. that Muslims feel discrimination blocking their integration into
a Western
identity and turn instead to an Islamic identity that supports
them in hostility
308 McCauley
toward those who will not accept them. The second explanation
is radical Islam:
the idea that religious fanaticism moves Muslims toward anti-
Western hostil-
ity. The third explanation is grievance, including both economic
and political
grievances.
These three explanations are not mutually contradictory, and it
is possible that
all three could contribute to radicalization of Muslims in
Western countries. Each
explanation may apply to the kind of radicalization that
produces active terrorists,
or to the mass radicalization that occurs in the pyramid of
sympathizers and
supporters that terrorists depend on, or to both. For instance,
economic frustration
may be associated with increased sympathy for terrorism in
opinion polls of
Muslims even if, as Krueger (2007) has argued, most Muslim
militants come from
families above average in socioeconomic status.
The present study focused on testing explanations of mass
radicalization of
U.S. Muslims. Only about one half of 1% of the U.S. population
30. are Muslims, a
small minority for which ordinary sampling methods are
prohibitively expensive.
Nevertheless, there are four national polls of U.S. Muslims
conducted between
2001 and 2007 with items relating to theories of radicalization.
Using these polls,
measures of socioeconomic status, religiosity, perceived
discrimination, and op-
position to U.S. forces in Afghanistan were correlated with
three elements of the
global-jihad frame: seeing the war on terrorism as a war on
Islam or “insincere,”
justifying suicide attacks in defense of Islam, and favorable
opinion of Al Qaeda.
The first result of interest is that the three aspects of the global-
jihad frame are
relatively independent in the one poll (Pew 2007) that included
all three relevant
items. Support for suicide attacks and favorable opinion of Al
Qaeda are correlated
(.30), but neither of these items is related to disapproval of the
war on terrorism.
This disjunction can be seen as well in the distribution of
opinions: few U.S.
Muslims favor suicide attacks or Al Qaeda, but most disapprove
of the U.S.-led
war on terrorism.
Disapproval of the war on terrorism is not associated with
gender, age, educa-
tion, family income, or religiosity, and is only weakly
associated with perception
of anti-Muslim bias in the United States. But Muslims who
oppose U.S. forces
31. in Afghanistan are much more likely to disapprove the war on
terrorism, and to
see it as a war on Islam. These results suggest that U.S.
Muslims evaluate the war
on terrorism more in terms of relations between the U.S. and
Muslim countries
than in terms of their own experience of discrimination in the
United States. If this
interpretation is correct, it follows that reducing discrimination
against Muslims in
the United States will not have much impact on U.S. Muslims
unless U.S. policies
change in relation to Muslim countries.
Compared with seeing a war on Islam, opinions favoring suicide
attacks and
Al Qaeda are relatively extreme: only 8% of Pew respondents
said that suicide
attacks are sometimes or often justified, and only 4% reported a
very or somewhat
favorable opinion of Al Qaeda. Although moderately correlated
(.30), these two
Testing Theories of Radicalization in Polls of U.S. Muslims 309
opinions have different predictors. More precisely, justifying
suicide attacks has
no useful predictors, with only slight tendencies for younger
and less-educated
respondents to report this rare opinion. But a favorable opinion
of Al Qaeda, no less
rare, is predicted by low income and opposition to U.S. forces
in Afghanistan. This
pattern of correlations was not predicted and, given that the Al
32. Qaeda and suicide-
attack items appeared only in the Pew 2007 poll, the pattern
should probably be
confirmed before any explanation is attempted.
Finally, in relation to the three explanations of radicalization
described in
the Introduction, the results of this study make a complex
pattern. Alienation, as
indexed by perceived discrimination, is consistently but weakly
associated with
seeing a war on Islam, but unrelated to justifying suicide
bombing or favorable
opinion of Al Qaeda. Radical Islam, as indexed by importance
of Islam and
frequency at mosque, is unrelated to any of the elements of the
jihadist frame.
Economic grievance, as indexed by low education or low family
income, is weakly
related to justifying suicide bombing and favorable opinion of
Al Qaeda. Political
grievance, represented by opposition to U.S. forces in
Afghanistan, is the best
predictor of a political judgment that the war on terrorism is
actually a war on
Islam and also predicts favorable opinion of Al Qaeda.
Overall, these results least support the idea that radicalization is
linked with
radical religion, and best support the idea that radicalization is
linked with opposi-
tion to U.S. policy in relation to Muslim countries. The
importance of opposition
to U.S. policy among U.S. Muslims is notably consistent with
the conclusions
of Tessler and Robbins (2007), who found that a negative
33. assessment of U.S.
foreign policy was a significant predictor of support for
terrorism in Algeria and
Jordan.
In brief, results of four polls indicate that different elements of
the global-
jihadist frame depend on different perceptions and experiences
of U.S. Muslims.
This kind of complexity is likely to increase as more is learned
about the political
opinions of U.S. Muslims: different theories may be required to
understand dif-
ferent aspects of opinion radicalization. There is also the
complexity represented
in different levels of the pyramid of radicalization (Moskalenko
& McCauley,
2009): theories of mass opinion radicalization may differ from
theories about
how individuals and groups make the transition to violent action
(McCauley &
Moskalenko, 2011).
Most generally, the results of this study suggest new directions
for the “war
of ideas” against terrorism. At least in the United States,
different elements of the
global-jihadist frame are believed by different subsets of
Muslims. These different
audiences will likely need different communications, perhaps in
different media,
to combat sympathy and support for jihadist terrorism. The war
of ideas is a
political competition between the United States and its enemies,
and the usual
tools of political competition—audience research, market
34. segmentation, targeted
communication—will be required.
310 McCauley
References
BBC News. (2010). Terrorism arrests rising in U.K. 25
February. Retrieved July 14, 2011, from
http://news.bbc.co.uk/2/hi/uk_news/8536412.stm.
Bejelopera, J. P., & Randall, M. A. (2010). American Jihadist
terrorism: Combating a complex
threat. Washington, DC: Congressional Research Service.
Retrieved July 14, 2011, from
http://www.fas.org/sgp/crs/terror/R41416.pdf.
Bergen, P., & Hoffman, B. (2010). Assessing the terrorist
threat. Bipartisan Policy Center.
Retrieved July 14, 2011, from
http://www.bipartisanpolicy.org/sites/default/files/NSPG%
20Final%20Threat%20Assessment.pdf.
Betz, D. (2008). The virtual dimension of contemporary
insurgency and counterinsurgency. Small
Wars and Insurgencies, 19(4), 510–540.
Blair, D. C. (2009). Director of National Intelligence: Annual
Threat Assessment of the Intelligence
Community for the Senate Select Committee on Intelligence.
Retrieved July 14, 2011, from
http://www.dni.gov/testimonies/20090212_testimony.pdf.
Brown, G. P., & Smith, J. (2009). Pursue, prevent, protect,
35. prepare: The United Kingdom’s strategy
for countering international terrorism. Crown Copyright,
corrected May 2009. Retrieved July
14, 2011, from http://www.official-
documents.gov.uk/document/cm78/7833/7833.pdf.
Dugan, L., Huang, J. Y., LaFree, G., & McCauley, C. (2009).
Sudden desistance from terrorism:
The Armenian Secret Army for the Liberation of Armenia and
the Justice Commandos of the
Armenian Genocide. Dynamics of Asymmetric Conflict, 1(3),
231–249.
Eikmeier, D. C. (2007). Qutbism: An ideology of Islamic-
Fascism. Parameters, (Spring), 85–98. Re-
trieved July 14, 2011, from
http://www.carlisle.army.mil/usawc/parameters/Articles/07sprin
g/
eikmeier.htm.
Fukuyama, F. (2006). Identity, immigration, and liberal
democracy. Journal of Democracy, 17(2),
5–20.
Johnson, T. (2007). The Taliban Insurgency and an Analysis of
Shabnamah (Night Letters). Small
Wars and Insurgencies, 18(3), 321–322.
JewsOnFirst. (2008). Insertion of millions of Obsession DVDs
in swing-state newspapers appears
to aid McCain campaign; Little-known Clarion Fund is spending
millions to distribute
controversial anti-Muslim video. Retrieved July 14, 2011, from
http://www.jewsonfirst.org/
08a/obsession.html.
36. Krueger, A. B. (2007). What makes a terrorist: Economics and
the roots of terrorism. Princeton, NJ:
Princeton University Press.
Leuprecht, C., Hataley, T., Moskalenko, S., & McCauley, C.
(2010). Containing the narrative: Strategy
and tactics in countering the storyline of global Jihad. Journal
of Policing, Intelligence and
Counter Terrorism, 5(1), 40–55.
Lloyd, J. (2003). Islam’s battle with a hostile world. Financial
Times. Retrieved July 14, 2011, from
http://www.danielpipes.org/996/islams-battle-with-a-hostile-
world.
McCauley, C., & Moskalenko, S. (2011). Friction: How
radicalization happens to them and us. New
York: Oxford University Press.
Moskalenko, S., & McCauley, C. (2009). Measuring political
mobilization: The distinction between
activism and radicalism. Terrorism and Political Violence,
21(2), 239–260.
Office of the Coordinator for Counterterrorism, U.S.
Department of State. (2009). Country Re-
ports on Terrorism 2008: Chapter 1. Strategic Assessment.
Retrieved July 14, 2011, from
http://www.state.gov/s/ct/rls/crt/2008/122411.htm.
Pew Research Center. (2007). Muslim Americans: Middle class
and mostly mainstream. Retrieved July
14, 2011, from http://pewresearch.org/pubs/483/muslim-
americans.
Roy, O. (2004). Globalized Islam: The search for a new
37. Ummah. New York: Columbia University
Press.
Sageman, M. (2004). Understanding terror networks.
Philadelphia: University of Pennsylvania Press.
Smart, C. L. (2005). The global war on terror: Mistaking
ideology as the center of gravity. Issue paper,
Center for Strategic Leadership, U.S. Army War College,
Volume 08-05, July. Retrieved July
14, 2011, from http://www.au.af.mil/au/awc/awcgate/army-
usawc/csl-gwot-cog.pdf.
Testing Theories of Radicalization in Polls of U.S. Muslims 311
Tessler, M., & Robbins, M. D. H. (2007). What leads some
ordinary Arab men and women to approve
of terrorist acts against the United States? Journal of Conflict
Resolution, 51(2), 305–328.
Wheatley, J., & McCauley, C. (2009). Losing your audience:
Desistance from terrorism in Egypt after
Luxor. Dynamics of Asymmetric Conflict, 1(3), 250–268.
CLARK McCAULEY is Rachel C. Hale Professor of Sciences
and Mathematics
and codirector of the Solomon Asch Center for Study of
Ethnopolitical Conflict
at Bryn Mawr College. He received his Ph.D. in social
psychology from the
University of Pennsylvania in 1970. He is coauthor of Why Not
Kill Them All?
The Logic and Prevention of Mass Political Murder (Princeton,
2006), coauthor
38. of Friction: How Radicalization Happens to Them and Us
(Oxford, 2011), and
founding editor of the journal Dynamics of Asymmetric
Conflict: Pathways toward
Terrorism and Genocide.
Copyright of Analyses of Social Issues & Public Policy is the
property of Wiley-Blackwell and its content may
not be copied or emailed to multiple sites or posted to a listserv
without the copyright holder's express written
permission. However, users may print, download, or email
articles for individual use.
Sampling Demystified
OVERVIEW
Planning research is an intricate dance between choosing the
research questions ,
identifying the likely measures and data sources, and
developing a sampling strategy
that fits with the data collection methods selected. While each
of the steps gets its
own chapter, the actual planning proq;~ss is not linear. Indeed,
it can be both messy
and frustrating because it goes round and round for a while.
This is normal. It takes
time .to get all the pieces aligned.
39. As the researchers consider the likely measures and data
collection strategies, they
also wrestle with how much data to collect. Is it is possible
review every file , observe
every street, or talk to every person in their class? Collecting
data from the whole
population of interest makes it possible to accurately report the
qualifications of every
teacher in the school system, the views of all the students in the
class, or the quality
of all the roads in the city. However, collecting data from every
file, street, or person
often requires too much time and money. Instead, researchers
opt to gather the data
from a subset of the entire population-that is, they work with a
sample.
This chapter explains sampling jargon, types of samples, and
how to determine the
right sample size. Sampling c6ncepts, such as sampling errors ,
confidence levels, and
confidence intervals are presented. Nonsampling errors are also
discussed.
Sampling has many benefits and is a marvel of statistical
wizardry. But there are
al so some challenges, not the least of which is an inherent
potential for error. When
you talk only to some people but not others , you can make
wrong predictions. The
classic example is the 1948 presidential election when President
Harry S. Truman
was challenged by Thomas E. Dewey. The Chicago Tribune
went to press before all
the otes were counted. Relying on polls that suggested the
inevitability of a Dewey
40. ictory, its headline read: "Dewey Defeats Truman." But the
polls were wrong, and
the next day Truman was declared the winner.
1be takeaway lesson is that random sampling can be very
powerful, but there is
- an inherent potential for error when working with sample data.
14 5
146 CHAPTER 10
SAMPLING JARGON
Sampling has some particular language that researchers and
sophisticated users of
research results need to know. Here are the primary terms and
concepts:
Population:
Census:
Sample:
Sampling
frame:
Sample
design:
Statistic:
41. Parameter:
Generaliz-
ability:
While population typically means all the people inhabiting a
specific
geographic area, the meaning is broader in social science. The
popula-
tion could be students at the university, news stories about the
2012
presidential candidates, the streets in the city, the streets in a
specific
neighborhood, or the applications for a specific zoning variance
in the
past year. I sometimes refer to this as the "population of
interest."
A count of the entire population. This could mean every student
in the
university, every news article about the 2012 presidential
candidates,
every street in the city, or every application for a zoning
variance in
the past year.
A subset of the population.
The list of all members of the population of interest, with the
specific
characteristics needed for the research. For example, the
sampling frame
could be all the students in a school, all the students in the
school who
are majoring in science and engineering, or all the
undergraduate students
42. who are majoring in science and engineering at all the state
four-year
colleges and university.
The method of sample selection. Most broadly, there are random
and
nonrandom sampling methods. Within each of these two large
catego-
ries, there are options.
A random sample means that every unit in the population of
inter-
est (whether files, streets, or people) has an equal chance of
being
selected.
A nonrandom sample means that files, streets, or people have
been
selected based on SOn;e criteria and do not have an equal
chance of
being selected. "
This very narrow meaning rs defined as a characteristic in the
population
based on sampling results. For example, the average salary of
the popula-
tion is estimated to be $35,000. This is also called the point
estimate.
This is the inference made about the likely range of the statistic.
For
example, the average salary in the population is estimated to be
between
$33,000 and $36,000. This is also called the confidence
interval.
43. This is a term used when working with data from a random
sample. It
means that based on the sample results, the researchers can
generalize
about the larger population of interest. Put another way, it is the
belief
that the random sample results fairly accurately represent what
is true
about the population of interest in general.
SAMPLIN G DEMYSTI FIED 147
RANDOM AND NONRANDOM SAMPLES
Samples are a subset of the population of interest-whether
people, files, salaman-
ders, roads, or bike lanes. There are two broad types: random
and nonrandom. There
are reasons for selecting the type of sample and each have
strengths and challenges.
Whether researchers use a random or nonrandom sample, they
should provide a
clear statement about the sample used and their rationale for
selection. It also helps
if the researchers gather and report demographic or other
essential information so
the readers can make an independent assessment about how
similar the sample is to
the population.
RANDOM SAMPLES
We are all familiar with blood tests . The lab workers do not
take all a patient's
44. blood. They take only a sample and assume it to be an accurate
reflection of all the
perso·n's blood . They then infer, draw conclusions, or make
generalizations about
the patient's health based on that sample. The idea of drawing a
sample for research
is the· same.
When scientists use the term "random," they do not mean
"whatever." It has a very
narrow definition that reflects its grounding in probability
theory. A random sample
means that every unit in the population ·of interest-whether it is
a population of
people, files, classrooms, or streets-has an equal chance of
being selected.
Randpm selection should not be confused with random
assignment used in experi-
mental designs. While both random selection and random
assignment serve to reduce
bias, in the language presented in Chapter 5, random assignment
enhances internal
validity while random selection enhances external validity. For
example, much of what
we know about social psychology comes from experiments
using college freshmen
randomly assigned to treatment or control groups. However,
those freshmen were not
randomly selected from the larger population and therefore are
not representative of
everyone in the population. Or using more jargon, the results
from college freshmen
are not generalizable to all adults .
In the world of science, if a sample is to be used, the ideal
45. choice is random. A
random sample enables researchers to do two very important
things. First, random
ampling minimizes selection bias because the selection of who
is included in the
tudy is taken out of the researchers' hands. Researchers cannot
select only those
people who look like them or who share a particular viewpoint.
They cannot choose
thinner files rather than thicker files. They cannot choose
streets where they live and
ignore streets in other parts of the city. A random sample should
be representative of
the population as a whole.
Second, random sampling enables researchers to make
mathematical estimates
ut the population based on the sample ' s results. The jargon
shorthand for this
ility to infer what is true about a population based on the results
of a random
sample is called generalizing the results. Researchers must use
random sampling
- y v ant to make general statements about the population as a
whole based on
· re ults.
148 CHAPTER 10
NONRANDOM SAMPLES
46. When a nonrandom sampling method is used, the magical power
of inference is lost.
A nonrandom sample, however, is frequently used because it is
the best option given
the situation (see Application 10.1 ). For example, if
researchers want tp understand
why a few communities are considered leaders in sustainable
growth, there is no in-
tention to generalize to all communities engaged in sustainable
growth. In this kind
of best practices research, a nonrandom sample of a few
communities recognized as
leaders in sustainable growth makes sense. The research
approach here is basically
a case study.
Nonrandom sampling is often preferable in situations where
only a small group
will be observed, interviewed, or surveyed. It is also used when
researchers do not
have all the information necessary for random sampling. For
example, researchers
are likely to talk with a small number of homeless people in
their community that
have been nonrandomly selected because the entire homeless
population is unknown.
Lastly, nonrandom sampling is useful when researchers are
doing exploratory work
to learn about the issues as the initial step in designing a much
larger study.
Nonrandom sampling is also used when there is no need to
generalize to a larger
population. Sometimes the researcher's goal is to provide a few
case examples to
illustrate a particular situation rather than trying to infer that
47. these examples are true
for the entire population of interest.
While the random and nonrandom sampling designs are
presented as if they are an
either/qr choice, in reality they can be used in combination with
different data col-
lection 'strategies within the same study. To make things more
complex , two different
sampling approaches can be combined. For example, we might
want to determine
the attitudes of students about working in public administration.
Researchers may
opt to select a handful of graduate schools within their state that
have public admin-
istration programs and then randomly select students who will
be asked to complete
a survey. With a combination of random and nonrandom
selection procedures, the
results might be more focused on a specific group that cannot be
reached through
just random sampling alone.
RANDOM SAMPLES: THE OPTIONS'
Random samples (also called probability samples) mean that
every unit in the popu-
lation has an equal chance of being selected. In order to select a
random sample, the
researchers need a complete listing of every unit in the
population, although there are
a few approaches that can be used when there is no complete
list. Random sampling
techniques range from simple to complex. This section will
focus on four categories:
simple, systematic, stratified, and cluster samples.
48. SIMPLE RANDOM SAMPLE
Researchers establish a sample size (discussed later in this
chapter) and then randomly
select that number of people, files, roads, whatever from the
population of interest.
SAMPLING DEMYSTIFIED
Application 1 0.1
Teenage Mothers on Welfare
149
Unmarried teenage mothers received particular attention in the
run-up to major welfare reform
in 1996. From a policy perspective, welfare reforms aimed at
unwed teen parents were thought
to make sense. If the laws could change their behavior, then
fewer teenagers wo·uld become
parents outside of marriage. Although they represent a
relatively small group on the welfare
rolls, teen mothers are likely to receive welfare benefits for a
number of years.
One relationship the researchers wanted to explore was whether
the availability of welfare
caused teenagers to have babies. In addition, the researchers
wanted to understand the teen-
agers' views on various welfare reforms targeted at teenage
mothers and what they saw as
the consequences of ending welfare for teenage mothers. Lastly,
the researchers wanted to
49. understand why the teens made the choice to have a baby and
their ideas about pregnancy
prevention for teenagers (Johnson 2001 ).
The first plan was to send surveys to a random sample of
teenage mothers on welfare.
However, neither a national listing of all teenage mothers on
welfare nor a listing of teenage
mothers within state welfare agencies existed. With no
reasonable way to do random sam-
pling , the researchers went to plan 8: a qualitative study using
focus groups with a nonrandom
sample.
A decision was also made to recruit participants from programs
that provide services to un-
wed parents or programs that serve a broader population, such
as Job Corps. The researchers
selected sites that would include a diverse set of teenage
mothers on welfare: whites, Hispan-
ics, African Americans, and Native Americans living in urban
or rural settings. The researchers
also wanted to capture respondents' views based on different
ages. Participant groups were
organized bY age: those under seventeen, those between
seventeen and nineteen, and those
in their early twenties who were teenagers when they became
parents.
Over ninety teenage mothers participated in a total of nine focus
groups. All the participants
had to agree to take part in the focus groups, and minors had to
have the permission of their
parent or guardian. This is one way to document informed
consent and ensure that all partici-
pants were there voluntarily.
50. Given the situation, a judgmental nonrandom sample made
sense to obtain the views from a
diverse group. While the results are not generalizable to all
teenage mothers in those locations
or the nation, this study provided insight into the policy debate
as seen from the perspective
of those who would be most affected ~Y the proposed reforms .
For example, researchers plan to interview fifty clients from the
local social service
agency. The researchers have a list of all 300 clients from the
past five years . They
assign each client a number from 1 to 300. To make their
selection, they use a ran-
dom number table, such as the one found on the National
Institute of Standards and
Technology's website (listed in Appendix C on page 306).
Picking any three-digit
column from the random number table, the researchers select
the first fifty numbers
that are from 1 to 300. Those fifty numbers are matched to the
client names and will
constitute the sample of clients to be interviewed.
Selecting phone numbers from a phone book using this
technique, however, is
not a random sample. It is considered a biased method because
some people do not
SAMPLING DEMYSTIFIED 151
To select a stratified sample, researchers must divide the
population into strata based
51. on some meaningful characteristic to answer their research
questions. The researchers
would have a listing of each unit within each stratum and then
draw a random sample.
In a way, the procedure is like drawing multiple simple random
samples.
Sometimes people confuse stratified random samples with a
judgmental sample.
The difference is that a judgmental nonrandom sample is just
that-nonrandom.
It is true that stratification is based on a judgment-that is, a
decision to divide
the population of CEOs into groups based on gender and race.
However, there is an
element of randomness in the selection along with choosing a
large enough sample
to do statistical analyses. In a nonrandom judgmental sample, it
is possible to make
the final selection randomly to eliminate bias, but this is rarely
done. The key point
is that the judgmental samples are more typically nonrandomly
selected but in either
case are so small that no inferences can be made.
There are two basic types of stratified samples: proportional and
disproportional.
While the descriptions presented below to illustrate these
approaches are relatively
simple, it is important to keep in mind that stratified samples
can get complex.
PROPORTIOt:-IAL STRATIFIED SAMPLE
The key feature of a proportional stratified sample is that the
52. sample maintains the
same proportion of the different groups as in the population. For
example, say there
are 800 male and 200 female senior managers in a large federal
agency. In a propor-
tional sample of 80 male and 20 female managers, the
proportion of men and women
remaihs exactly the same as it is in the population (see Exhibit
10.1 ). This makes
sense when the intention is to obtain a sample that will give a
reasonable estimate of
the population as a whole. However, it does not make sense if
the researchers want
to make statistical comparisons between the men and women.
D ISPROPORTIONATE STRATIFIED SAMPLE
If the researchers want to make comparisons between the men
and women, they
will use a disproportionate stratified sample. This means that
the researchers will
deliberately randomly select a larger number from the smaller
groups relative to their
proportion in the population. --this is disproportionate
sampling.
Suppose the researchers want a 50/50 split between the male
and female managers
in the federal agency in their study. The sample size is still 100,
so the researchers
will randomly select 50 men and 50 women (see Exhibit 10.2).
In the jargon, the
researchers will oversample the women.
The stratification can get complex. For example, if the
researchers want to compare
53. the views of white men, minority men, white women, and
minority women, they
would need to have four strata. However, if the number of all
minority men, white
women, and/or minority women were small, the researchers
might select all of them
and then select a random sample of the white male managers in
the federal agency.
This would still be considered a disproportionate stratified
sample.
When the U.S. GovernmentAccountability Office (1990a)
decided to survey busi-
nesses across the country to determine the extent of
discrimination related to the Im-
152
Exhibit 1 0.1 Senior Managers in a Federal Agency:
Proportional Sample
Men
Women
Total
Population
800
200
1,000
Percent
distribution Sample
54. 80 80
20 20
100 100
Exhibit 10.2 Senior Managers in a Federal Agency:
Disproportionate Sample
Percent
Population distribution Sample
Men 800 80 50
Women 200 20 50
Total 1,000 100 100
CHAPTER 10
Percent
distribution
80
20
100
Percent
distribution
50
50
100
migration Reform and Control Act, it chose a disproportionate
random sample design.
55. The· rationale was that the opportunity to discriminate against
people who look and
sound foreign would vary based on the industry and area of the
country. Given these
factors , almost 10,000 businesses were randomly selected with
a disproportionate
technique using fifty-six strata. Talk about complex! But it was
necessary so different
industries in different areas of the country could be compared.
Analyzing Disproportionate Samples
Having sufficient numbers in each of the strata makes sense so
that the results can
be compared. However, this creates some problems when
generalizing to the larger
population. The stratum that was oversampled distorts the
aggregate results unless it
is corrected in the analysis. Disproportionate stratified samples
need to be weighted
so that the proportions of the strata in the sample are the same
as the proportions in
the population.
Let us look at a dramatic example of the results of a
disproportionate sample when
it is unweighted and weighted . Thes~ are fake data; it is
unlikely that real life would
have anything so dramatic.
Researchers have taken a disproportionate stratified random
sample of 400 graduate
students from a population of 4 ,000. They also selected 400
undergraduates from a
population of 16,000 students. One survey question asks how
satisfied or dissatisfied
56. the students are with the university. The results are displayed in
Exhibit 1 0.3 . Ifthe
researchers are only going to refer to the separate strata and not
combine them, they
could accurately state that I 00 percent of the undergraduates
were satisfied while 100
percent of the graduate students were dissatisfied.
However, if the researchers wanted to give an overall picture of
all students at the
university, the unweighted summary would be misleading.
Based on the unweighted
data, the researchers would conclude that half of all the students
are satisfied. This is
SAMPLING DEMYSTIFIED 153
Exhibit 10.3 Satisfaction with School: Analysis with
Unweighted Data
Graduate Undergraduate
students students Total
Satisfied 0% 100% 50%
n=400
Dissatisfied 100% 0% 50%
n=400
Total n= 400 n= 400 n= BOO
100%
Exhibit 10.4 Satisfaction with School: Analysis with Weighted
Data
57. Graduate Undergraduate
students students Total
Satisfied 0 100% 80%
N= 16,000
Dissatisfied 100% 0 20%
N= 4,000
Total N= 20,000
because the unweighted sample has a disproportionate number
of graduate students
relative to the population of students arid their dissatisfaction
distorts the generalized
results to all students.
Weighting the data means putting the groups back into the
proportion as they
exist in the population. Each graduate student in the sample
represents ten students
(calculated: 4,000/400). Each undergraduate in the sample
represents forty students
(calculated: 16,000/400). Multiplying the 400 graduate students
who reported being
dissatisfied by 10 totals 4,000. For the undergraduates,
multiplying the 400 respon-
dents who report being satisfied by 40 totals 16,000.
The correctly weighted data show that 80 percent of all students
report being
satisfied as compared to 20 percent who report being
dissatisfied (see Exhibit 10.4 ).
If this were real data, the university officials should be
concerned about the views
58. of the graduate students, but when generalizing about all
students, it is accurate to
report an 80 percent satisfaction le¥el.
The takeaway lesson is that the weighted and unweighted results
can be dramati-
cally different. Therefore, researchers must use weighted data to
generalize results
to the whole population if they are working with a
disproportionate stratified random
sample. Researchers might show the weighted and unweighted
results for informa-
tional purposes, but they have to show the weighted data if they
are making any kind
of generalization about the population from which the sample
was selected.
C USTER SAMPLE
Cluster sampling is a multistep approach to select a random
sample when a complete
listi ng of the population does not exist (see Application 10.2).
It is particularly use-
SAMPLING DEMYSTIF IED 155
ful when working over a large geographic area. For example,
researchers who want
to find out MPA students ' future plans after graduation will
quickly find that there
is no national list of all students. To locate students, the
researchers would have to
contact every program in the country. However, if there are not
enough resources
59. to contact every program but they still want to do a random
sample, they will use a
cluster sampling approach.
To do a cluster sample, the researchers would first obtain a
listing of all MPA
programs in the country. They would randomly select maybe
twenty programs and
contact them in order to obtain the names of all the students. It
is unlikely, however,
that the schools would give out the names because of privacy
laws. The school officials
might, however, let the researchers administer surveys in
classes. The researchers
would randomly select classes likely to be attended by those
closest to graduation.
Or you might think of this in terms of rounds. Round 1: Obtain
list of all MPA
programs. Round 2: Randomly select a small number of schools.
Round 3: Randomly
select MPA classes and administer survey to all students in
those classes.
' One drawback of the initial selection is that the twenty schools
may not reflect all
MPA programs or all the students because it is a relatively
small number. While random
selection reduces research bias in selection, it is possible to
wind up with schools all
located in large metropolitan areas on the east and west coasts.
The study 's results
might not be representative of all MPA students. However, it
may be a limitation that
the researchers are willing to live with. given the resources
available. This is the kind
60. of limitation that must be stated in the reported results and the
conclusions.
Alternatively, the researchers could make the initial school
selections judgmentally.
For example, they might choose five schools from different
regions of the country:
the northwest, southwest, northeast, southeast, north central,
and south central. One
takeaway lesson is that it is possible to combine elements of
random and nonrandom
selection within a single study.
Generalizing back to the population, however, is more
problematic because of the
lack of an initial random selection. The researchers should not
attempt to generalize
back to the larger population unless there is randomness
throughout. Still, sometimes
there is no need to make statements about all MPA students.
There may be enough
information based on what the researchers learned from these
students from these
particular schools to provid€ 'some useful insights. It also
provides a guide for repli-
cation; other researchers could do,the same research using
different schools. If there
begins to be a similar pattern, the results gain credibility. If
enough studies are done,
a researcher can pull them all together in a meta-analysis .
NONRANDOM SAMPLES: THE OPTIONS
Researchers have several choices in nonrandom sampling
methods: quota, judgmental,
accidental, convenience, and snowball.
61. Researchers use a quota sample to make sure specific numbers
of members of
different subgroups are included. For example, researchers
decide to interview ten
women and ten men at a health clinic. They approach men and
women and accept the
first ten in each group who agree to participate in the interviews
. Another example is
156 CHAPTER 10
a telephone survey that is designed to include fifty Republicans
and fifty Democrats.
The researchers call until they have filled the quota; if they get
fifty Republicans fairly
easily, they must keep dialing for Democrats until they get fifty.
A judgmental sample, which is sometimes called a purposive
sample, shares the
goal of ensuring that the different subgroups of the population
are incl4ded. However,
selections are based on very specific, predetermined criteria that
make sense given
the research question and the situation. For example, if
researchers are interested
in understanding why some elementary schools are successful as
measured by high
student scores on standardized tests, they may decide to
interview the principals of
the schools with the five highest scores. How is this different
from a quota sample? A
quota sample would accept the first five principals who raise
their hands at a confer-
62. ence to volunteer to participate. In a judgmental sample, the
principals are selected
on specific criteria and are named.
An accidental sample is the classic "person on the street"
sample, sort of like the
Tonight Show's "Jay-walking" although probably not as funny.
Accidental does not
mean random in the narrow statistical sense. If researchers
stand outside one super-
market on Saturday to ask people to fill out a survey, only those
shopping at that time
and place have a chance of being selected. Not every person in
the community has
an equal chance of being selected, thus violating the definition
of random sample:
every unit of the population of interest must have an equal
chance of being selected.
A reasonable assumption is that the ~ople shopping at this time
and place are not
representative of the community as a whole.
A ~onvenience sample is similar to an accidental sample in that
not everyone has an
equaf chance of being selected. For example, it is convenient
for an MPA professor to
ask her students to complete a survey about public service
motivation. However, these
students are not representative of all public service employees,
of all MPA students
across the country, or even of all students in that particular
MPA program. As long
as the reported results stay within the limitations of the
sampling approach used, the
survey results can be useful. This type of sampling approach is
a good starting place
63. for exploratory research.
A snowball sample is used when researchers do not know whom
to include and rely
on others to tell them. For example, if researchers are trying to
determine the informal
power structure in an unfamiliar community, they probably will
not know with whom
they should talk. The researchers would begin with the people
most likely to have the
desired information about power in the community. At the end
of the interview, they
will ask, "Whom else should I talk with?" Ideally, they would
continue this snowball
process until no new names are suggested. At that point, they
will feel fairly certain
that they interviewed all the key people knowledgeable about
this issue.
DETERMINING SAMPLE SIZE
Sample size is not a concern when working with nonrandom
samples. Whether they
are large or small, they are never "generalizable" to the larger
population. However,
sample size is a huge issue for random samples, which need to
be large enough so
researchers can determine how likely the results are a fairly
accurate reflection of the
SAMPU G DEMYSTIFIED 157
population. How do researchers decide on how large a random
sample should be? The
64. decision is based on three factors. One is the size of the
population, which is usually
known . For example, we probably know how many people are
in the state, how many
students are in the classroom, or how many streets are in a
defined neighborhood.
Sometimes, however, this information is not known. For
example, the researchers
may not know exactly how many files are in the boxes stored in
the basement. But
for my purposes here, we will assume that the population size is
known.
The second factor is how sure the researchers want to be that
sample results are
a fairly accurate representation of the population; this is called
the confidence level.
The social science standard is to set the confidence level at 95
percent. In the world
of statistics , this translates to a statement that says: If the
researchers selected 100
different samples , only 5 percent would be very different from
the population. Put
more simply, this means that the sample results have only a 5
percent chance of be-
ing wrong. Or put still another way, the researchers are 95
percent sure that if they
had surveyed everyone in the population , the results would
have been fairly similar.
Those are good odds .
There is a relationship between the sample size and the
confidence level. If re-
searchers drop to a 90 percent confidence level, there is a 10
percent chance of be-
ing wrong . The needed sample size will be smaller than at the
65. 95 percent level. If
researchers want to be 99 percent confident, meaning that there
is only a 1 percent
chance of being wrong, researchers will need a larger sample.
There is a trade-off
between sample size and the amount of risk of having a sample
that is not a good
repres~tntation of the population.
The third factor in deciding the size of the sample is precision:
how precise do the
researchers want to be in making estimates about the population
based on the sample
results? Precision is sometimes called "sampling error," "the
margin of error," or "the
confidence interval."
For example, polling data are usually reported in terms of the
margin of error.
National polls typically set their standard at plus or minus 3
percent. A poll might
report that 49 percent plan to vote for the Republican candidate
and 51 percent for the
Democratic candidate, with a sampling error of plus or minus 3
percent. In statistics
language, this means that the pollsters are 95 percent certain
that if they had surveyed
everyone, the true likely vote tor the Republican candidate
would be between 52
percent (49 percent plus 3) and 46 percent (49 percent minus 3).
Similarly, the true
likely vote for the Democrat would be between 54 percent and
48 percent (51 percent
plus 3 and 51 percent minus 3). When the sampling errors
overlap, as they do in thi s
fake example, it means that the election is too close to call.
66. When working with real numbers (interval and ratio level data),
the precision is
presented in terms of the confidence interval (not to be
confused with the confidence
level, explained above; this is yet another example of similar
terms with different
meanings that make us all crazy). Researchers use confidence
intervals when they want
to estimate the range of the actual number or average in the
population based on the
sample results-in sampling jargon, when they want to estimate
the parameters.
For example, if the average salary in a sample is $20,000, this
becomes the point
estimate in the population. The computer then calculates the
confidence interval; it is
158 CHAPTER l 0
Exhibit 10.5 Guide to Sample Size
Population size Sample size
10 10
50 44
100 80
200 132
250 152
67. 500 217
700 248
1,000 278
3,000 341
50,000 381
100,000+ 384
Source: Krejcie and Morgan , 1970.
Note : These are based on the social science standard of 5
percent precision and 95 percent confidence
level. ·
$18,000 at the low end and $22,000 at the high end. The
researchers would state that
they are 95 percent certain (this is the confidence level) that the
true average salary of
the population is between $18,000 and $22,000 (this is the
confidence interval). In the
Iraqi mohality study discussed in Application 1 0.2, the Lancet
researchers reported that
they were 95 percent certain that the true number of deaths was
somewhere between
393,000 and 943,000; the researchers are reporting the
confidence interval.
The key point is that sample size is affected by the choices of
desired precision
and confidence levels. Statisticians have developed the formula
and tested the math
68. to enable researchers to choose the right sample size depending
on how much risk
they want to take (see Exhibit 1 0.5). At the social science
standard of 5 percent pre-
cision and a confidence level of 95 percent, the researchers need
a sample of 384 if
the population is 100,000 or greater. However, if the
researchers wanted to be very
precise at the plus or minus 1 percent, they would need a sample
size of 9,000 for a
population of 100,000.
It is also important to note that the smaller the population, the
higher the propor-
tion of cases need to be selected in the random sample. If the
population is 200, the
sample size will be 132. At 700, the sample will be 248. When
the population is
larger than 100,000, 384 are needed. These are based on a 5
percent precision and a
95 percent confidence level.
The last factor in calculating sample size is the proportion of
the characteristic in
the population of interest. The population proportion means that
if the researchers
were looking at the gender variable, they would assume there
would be a 50/50 split
between men and women. The most conservative estimate is .50,
meaning that the
characteristic is found in 50 percent of the population. It is
considered a conservative
measure because it results in the largest sample size to meet the
social science stan-
69. SAMPLING DEMYSTIFIED 159
dards of 5 percent precision and a 95 percent confidence level
(Krejcie and Morgan
1970). If the researchers were considering a population
consisting of managers and
nonmanagers in an organization, the population proportion
might be 1 0 percent man-
agers and 90 percent nonmanagers. Statisticians would calculate
the precise sample
size based on all these variables using the formula in Appendix
A (see page 277), but
the sample sizes in Exhibit 10.5 provide useful conservative
estimates.
It is important to remember that these sizes are minimums. A
researcher may want
to select a sample size somewhat larger if the data collection is
not likely to receive a
100 percent response rate. In order to compare strata,
researchers should select sample
sizes for each stratum based on the population and desired
degree of confidence
and precision so they can obtain statistically valid comparisons.
For example, if the
researchers want to do in-depth analyses of differences in
viewpoints of managers
and nonmanagers in the organization, they would use a
stratified random sample to
ensure that they have enough managers to do meaningful
comparisons.
While samples are used to keep down the costs of data
collection, my advice to
any researcher is to try to gather data from the whole population
70. if it is feasible. If an
organization has 100 employees, it is best to survey them all
rather than just eighty.
However, if the population is just too large and a sample is
needed, researchers should
select as large a sample as they can manage. There is nothing
more frustrating than
doing all the work involved in research only to be unable to
draw conclusions or
compare the responses of groups within the organization
because the sample was too
small . Do not skimp on sampling!
NONSAMPLI N G ERRORS
Research results may have errors that have nothing to do with
the probabilities of
error associated with random samples. These nonsampling
errors refer to possible
problems with how the data were collected. A major
nonsampling error is a low re-
sponse rate. Even when the researchers select a large enough
random sample, if few
people respond, the researchers are left with a volunteer sample:
those who chose to
participate in the survey might be different from the rest of the
population.
Other nonsampling errors can ·come from the questions and
responses. When in-
terviewers ask different people s lightly differently questions,
the data are no longer
reliable and create a nonsampling error. Nonsampling errors can
also occur when
people do not give truthful answers. For example, people may
say what they believe
71. the researcher wants to hear or what they think is the socially
desirable response.
"Of course," someone might say to a woman interviewer, "I
would vote for a quali-
fied woman for president." Whether that is a truthful answer,
however, is unknown .
Asking if people use illegal drugs runs the risk of their saying
"no" because they
believe it is safer.
Researchers should do what they can to minimize nonsampling
error and acknowl-
edge it as a possible limitation of their study. For example, in
the Iraq mortality study
(Burnham et al. 2006), the researchers attempted to minimize
nonsampling errors in
a variety of ways . In order to minimize response bias they had
equal numbers of men
and women on the teams. All interviewers were medical
doctors, fluent in Arabic
Running Head: RESEARCH QUESTION AND EVALUATION
DESIGN
RESEARCH QUESTION AND EVALUATION DESIGN
Research Question and Evaluation Design
Name
Instructor
Date
Introduction
Bata is an international shoe making company with branches all
over the world. Its headquarters is in Central Michigan, and it
72. was begun in the year 1902. Since then it has expanded greatly
and gained popularity worldwide. To increase its sells even
more, it engaged in a research and evaluation program to
determine which types of shoes are bestsold, and also to prepare
adequately to ensure that they produce the best products to suit
every season. The program manager came up with a team to
undertake the research and to ensure that the best results are
obtained (Johnson, 2014).
Activities
To ensure that they acquire the best results, there was a lot to be
done in preparation for the filed study, which included
conducting a reconnaissance and preparing the research
equipment tools.
Methods of research
1. Observation
This is a primary way of obtaining first hand research
information. It involves getting into the filed with aim to
observe and record the results for evaluation. This was
considered the most important and most basic way to identify
the most preferred type of shoe. The team divided itself and
travelled to several parts of different towns and states. The
results of all types of shoes worn were recorded down in a
tallying table. When the collection of information was done, the
analysis followed whereby the means of common tendency
which are the mean, mode and the media, were used to analyze
the data (Kawulich, May,2005).
2. Questionnaire
This was the most tedious of the methods applied in the
research. It involved preparation of questions in booklets called
73. questionnaires. A great care was taken to ensure that the
questions posted do not touch into the privacy of the
respondents. The questions also should be clear with no
ambiguity. The exercise took two forms whereby some
questionnaireswere posted to theidentified respondents which
were meant to save on travelling costs and time. Another way
was taking the questionnaires to individual respondents where
they filled the forms and were returned soon after finishing.
This method had the advantage in that lots of similar questions
were given to different people hence many answers came out
which helped greatly in comparing the customers ‘preferences.
(Qualities of a good question., 2009).
3. Interviewing
The last method of research was through interviews. The team
brought forth a number of questions which included asking the
respondent the best type of shoe they loves, whether rubber or
leather, the shoe number, the durability of the shoe and the
suitability of the shoe type according to each season. This
method was likely to bring good results because of the face to
face interaction of the respondent and the researcher. This
enabled the researcher to dispel out any anxiety or fear from the
respondent to ensure they give out responses which are out of
free will. It also enabled the researcher to ask
supplementaryquestions to unclear responses so as to obtain
clear answers to the research. Notes were taken which were
later analyzed to get the desired feedback (Bill, 1st August,
2005)
Results analysis
After the extensive research, the researchers had the task of
analyzing and interpreting the results for the purpose of
presenting to the management. From the record tallies,
observations and the questionnaires, the most preferred shoe
type of shoe is the leather shoe of average size that is six, seven
74. and eight.
Conclusion
After the presentation of the results, the Bata Company was in a
position to prepare adequately to meet the customers’ demands.
The new direction to be taken was now clear which was to
increase the production of the medium sized leather shoes.
(Designing an experiment)
References
Qualities of a good question. (2009). Retrieved from StatPac:
http://www.statpac.com/surveys/question-qualities.htm
Bill, G. (1st August, 2005). Research Interviewing: The Range
Of Techniques: A Practical Guide. McGraw-Hill International.
(n.d.). Designing an experiment.
Johnson, G. (2014). Research methods for public administrators
(3rd ed.). NewYork: Armonk.
Kawulich, B. B. (May,2005). Participant Observation as a Data
Collection Method. Carrollton.