Skip to content

Predicting Justification of Suicide Bombing to Defend Islam

May 25, 2010

MALCOLM L. RIGSBY
Texas Woman’s University
13 May, 2010

Estimates Predicting Justification of Suicide Bombing to Defend Islam, U.S. Muslims Age 18 and Over, 2007

Ordinal Logistic Regression.  Ordinal Logistic Regression, also called Ordinal Regression was conducted to predict justification of suicide bombing to defend Islam in the United States among adult Muslim Americans aged 18 and over.  Table 1 presents Ordinal Regression logged odds estimates (B) predicting justification of suicide bombing to defend Islam in the United States, U.S. Muslims Age 18 and Over, 2007.  This data was retrieved from the 2007 Pew “Muslim American Survey” (N = 1050, Valid N = 825).  This variable appeared in the cited survey as question qh1 and was recoded where 1 = “never justified”, 2 = “rarely justified”, 3 = “sometimes justified”, and 4 = “often justified”.  As noted in the SPSS frequencies for this variable the cases were highly skewed where 87.7% of the cases reported “never justified” as the response to this question (Appendix I, page 4).   Ordinal regression was performed to estimate this variable upon six predictors.  The six predictors include Age, Sex (which was dummy coded 1 = Male, 0 = Female), level of education (an ordinal level variable with seven categories), Race  (a nominal level variable with four categories, which was recoded using White as the reference group), and Marriage Status which was nominal level with four categories of married, divorced, widowed, and separated).  This variable was recoded to two categories where 0 = not married and 1 = married.  The last variable selected was Income (an ordinal level variable with nine categories).  Recoding was conducted to eliminate answers of don’t know, refused, or missing.  

Examination of the Model Fit statistics in Table 1, reflect a highly significant -2 log likelihood (679) and Model χ2 (38) both of which are highly significant (p = .000, df = 8).  Accordingly this is a good model.  Also, regression generated a Pseudo R2 of .073 using Nagelkerke.  This is a low value and means the model explains only about 7% of predicting justification of suicide bombing to defend Islam among U.S. Muslims aged 18 and over in the United States in this data.

____________________________

Table 1 About Here
____________________________

 An examination of the model reflects that only two predictors have a significant relationship with use of suicide bombing to defend Islam in the United States as among Muslim American adults.  An examination of each predictor reflects the following.  Age has a negative and highly significant relationship to predicting justification of suicide bombing to defend Islam.  Specifically, the older a person is the less likely they are to justify use of suicide bombing to defend Islam (B = -.515, (.116), p ≤ .001).  Said another way, for each one year increase in age the probability of finding use of suicide bombing justified in defending Islam will decrease by 40% (℮-.515 = .598 – 1 =.402 = 40%).  This is logical in consideration that older people get the more conservative they may become.  Education also has a negative and significant relationship to predicting justified use of suicide bombing (B = -.152, (.070), p ≤ .05).  The more education a respondent has the more likely they are to not justify suicide bombing to defend suicide bombing.  In short for each level increase in education the probability of supporting use of suicide bombing to defend Islam decreases by 14% (℮-.320 = .859 – 1 =.141 = 14%).  Sex (being male) and other race each have a negative, but not significant relationship to predicting justification of suicide bombing to defend Islam.  Being Black, Asian, being married, and income level each have a positive, but insignificant relationship with justification to using suicide bombing (p > .05).[1]  

It is important to note that in conducting the ordinal regression that the assumption of parallel lines was tested using Negative Log-log and logit link functions.  Negative log-log was used in this presentation because in the data the lower categories tended to be the more likely responses.  In each case the null hypothesis that the location parameters (slope coefficients) are the same across response categories failed to be rejected at the .05 level.  Hence, as noted in the table 1, ordinal regression is the proper regression technique to perform in this analysis.

Factor Scale and Ordinal Regression.  Next in order to examine these findings more closely a factor scale was created represent an additional predictor for social economic status (SES_Factor) of respondents.  This exercise demonstrates the use of factor analysis to represent the relationships among a set of observed variables in terms of two common factors after which regression was conducted.  This was an exploratory analysis seeking to construct a latent variable as a factor representing social economic status, that could then be used in regression as a dependent variable and predictors from Table 1 to predict attitude toward violence.  The observed indicator variables used in the factor analysis phase were education_recoded as noted earlier; employment status (employ_recode); and income recoded as Incomesum2.  All three were ordinal level data.   Factor analysis technique was used to create a factor scale.  Table 2, below shows the one common factor identified, and factor loading generated in regard to it from using this technique (Eigenvalue Total = 1.715).  Eigenvalue of 1 or higher was used in this technique, hence the factor created is a good factor.  This value represents the correlation (the standardized values) between the observed indicators (education, employ, and income) as they co-vary between each other in relation to the factor loading.  As an example, Table 1 reflects that the correlation between indicator 1 (education) and SES_Factor is .799.  This means that predicting justification of suicide bombing to defend Islam loaded high in correlation to the Social Economic Status factor for attitude justifying suicide bombing to defend Islam.  Hence, the observed indicators are a linear function of the factor extracted from the analysis. The communality values explain the amount of variance in each observed indicator row explained by the SES_Factor.  Therefore when examining the communality value for indicator one it may be concluded that the SES_Factor explains about 64% of variance in predicting a respondent’s justification of suicide bombing to defend Islam, as among U.S. Muslims aged 18 and over in 2007.

____________________________

Table 2 About Here
____________________________

 Turning to Table 2 since only one factor was generated no factor rotation was calculated.  The SES_Factor explains about 57% of the covariance between the indicators used.   When it comes to level of education, the SES_Factor explains about 63% of variance in predicting justification of suicide bombing.  Likewise this factor explains about 43% of variance in education level.  SES has a positive, moderate correlation to income. 

            Ordinal Regression was then conducted in which SES_Factor was used as a predictor variable in a separate model.  The predictor variables selected for regression were the same as in the earlier ordinal regression presented in Table 1 above, except predictors “income” and “education” were dropped and replaced with SES_Factor.  The findings from regression with the SES_Factor (identified as Ordinal Model 2) are presented in Table 3 along with the findings from the earlier regression (identified as Ordinal Model 1) without the factor.

The Test of Parallel Lines in model 2 indicated an insignificant value (p > .05) therefore calling for failure to reject the null hypothesis.  Accordingly the assumption of parallel lines was met and Ordinal regression is proper to use in the model with the SES_Factor.   The SPSS report is attached as Appendix 2.

____________________________

Table 3 About Here
____________________________

 Examination of the Model Fit statistics for Model 2 in Table 3, reflect a -2 log likelihood (712) and Model χ2 (38) both of which are highly significant (p = .000, df = 8).  Accordingly this is a good model.  Also, regression generated a Pseudo R2 of .073 using Nagelkerke.  This is a low value and means the model explains only about 7% of predicting justification of suicide bombing to defend Islam among U.S. Muslims aged 18 and over in the United States in this data.  Comparing the model fit statistics of each model it is concluded that Model 1 which did not include the factor is better based upon the smaller -2 log likelihood value.  However it is important to note that the models are equal in predicting justification of suicide bombing to defend Islam and the Model χ2 of each model is the same value.

            As in Model 1 sex and other race remain negative and insignificant in relation to predicting justification of suicide bombing.  Likewise, Black, Asian, and marital status remain positive and insignificant in this relationship.  Age remains positive in Model 2 and highly significant (B = -.508, (.115), p ≤ .001).  This provides evidence to confirm that the older a person is the less likely they are to justify use of suicide bombing to defend Islam.  In this model, for each one year increase in age the probability of finding use of suicide bombing justified in defending Islam will decrease by 40% (℮-.508 = .602 – 1 =.398 = 40%).  Social economic status also has a negative and significant relationship to predicting justified use of suicide bombing (B = -.264, (.108), p ≤ .05).  The higher level of SES a respondent has the more likely they are not to justify suicide bombing to defend suicide bombing.  In short for each level increase in SES the probability of supporting use of suicide bombing to defend Islam decreases by 23% (℮-.264 = .768 – 1 =.232 = 23%).

            This evaluation was most interesting.  It allowed for a comparison and attempt to explore some data that is constantly being updated in relation to my primary interest in violence, politics, culture and religion’s seemingly interwoven role. 

Table 1.  Ordinal Regression Estimates Predicting Justification of Suicide Bombing to Defend Islam , U.S. Muslims Age 18 and Over, 2007

Predictor  

Ordinal Model

B

Threshold 1 -.059

(.425)

Threshold 2 -.412

(.430)

Threshold 3 1.978***

(.489)

Age -.515***

(.116)

Male -.327

(.207)

Education -.152**

(.070)

Race (ref=white)

Black dummy

Asian dummy

Other race dummy

 

.278

(.271)

.039

(.263)

-.237

(.337)

Marital status .092

(.235)

Income .005

(.050)

-2 log likelihood

Model χ2

Pseudo R2**

Degrees of freedom

N

                                     679

                                       38

                                     .073

                                         8

                                     825

*p ≤ .05   **p ≤ .01   ***p ≤ .001
** measured by Nagelkerke
Note:  Standard errors are in parentheses.  For ordinal regression, the assumption that the location parameters (slope coefficients) are the same across response categories is not met.  Testing of parallel lines using Negative Log-log, and Probit failed to reject the null hypothesis (p > .05).  Hence, there is support that the assumption of parallel lines is violated and therefore Ordinal regression is appropriate to examine the data.

Source:  Pew 2007 Muslim American Survey

Table 2:  Factor Loadings of Rotated Factors on Justification of Suicide Bombing to Defend Islam, U.S. Muslims Age 18 and Over, 2007. (N = 861)

  Factor loadings  

Communality

Variable F1

(SES)

hj2
  1. Education
.779 .638
  1. Employment
.652 .425
  1. Income
.808 .652
Eigenvalues 1.715  
% of variance explained 57.177  

 

Notes:  Extraction method:  principal components analysis
Rotation method:  Varimax with Keiser normalization

Table 3:  Ordinal Regression Estimates Predicting Justification of Suicide Bombing to Defend Islam, U.S. Muslims Age 18 and Over, 2007.  (N = 825)

  Ordinal Model 1 Ordinal Model 2
Predictor B B
Threshold 1 -.059

(.425)

                .775**

               (.342)

Threshold 2 -.412

(.430)

              1.247***

               (.349)

Threshold 3 1.978***

(.489)

              2.812***

               (.420)

Age -.515***

(.116)

               -.508***

               (.115)

Male -.327

(.207)

               -.266

               (.209)

Education -.152**

(.070)

 
Race (ref=white)

Black dummy

Asian dummy

Other race dummy

 

.278

(.271)

.039

(.263)

-.237

(.337)

 

                .268

               (.280)

                .028

               (.263)

               -.267

               (.336)

Marital status .092

(.235)

               .144

              (.235)

Income .005

(.050)

 
SES Factor                 -.264**

              (.108)

-2 log likelihood

Model χ2

Pseudo R2

Degrees of freedom

N

  679

  38

  .073

  8

  825

              712

              38

              .073

              7

              820

*p ≤ .05   **p ≤ .01   ***p ≤ .001
Note:  Standard errors are in parentheses.
Source:  Pew 2007 Muslim American Survey


[1] As a side observation, it is of great interest to me that being male rather than female fails to make a difference in justifying use of suicide bombing to defend Islam.  In an earlier regression using the same data set and the dependent variable “concern of rise of Islamic extremism in the United States as expressed by Muslim American adults there was a negative and significant relationship between being male as opposed to female and concern of rise of Islamic extremism in the U.S. [B = -.320, (.076), p ≤ .001].  From these findings it appears that Muslim American adult men are less likely to have more concern of rise of Islamic extremism in the United States than are women.  Converting the logged odds of -.320 to odds ratio it may be stated that Muslim American adult men are about 27% less likely to have a greater concern of rise of Islamic extremism than Muslim American adult women (℮-.320 = .726 – 1 =.273 = 27%).

Advertisements
No comments yet

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: