Predicting Attitudes Disapproving Use of Violence in Mulsim Americans, Aged 18 and Over
MALCOLM L. RIGSBY
Texas Woman’s University
15 April, 2010
Attitudes Disapproving Use Of Violence, U.S. Muslims
Age 18 and Over, 2007
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. All data was retrieved through the Pew 2007 Muslim American Survey. This was an exploratory analysis seeking to construct latent variables as factors representing attitude toward violence that could then be used in regression as dependent variables and using predictors to predict attitude toward violence. Initially, in conducting factor analysis, the observed indicator variables used in factor analysis were Concern For Extremism, an ordinal level variable recoded as ConcExtreemism with category 1 “very concerned” and 4 being “not at all concerned”; When Suicide Justified, also an ordinal level data recoded as JustiSuicide where 1 is “often justified” and 4 is “never justified”; Opinion of al Qaida an ordinal level variable recoded as “opinAlQaida” where 1 is “very favorable” and 4 is “very unfavorable”; and the fourth variable is Opinion of Afghan War Decision which is recoded as AfghanWarDecision. This variable was dichotomous with 1 = right decision and 2 = wrong decision. Factor analysis technique was used to create a factor scale. Table 1, below shows the two factors and factor loading generated in regard to each from using this technique. The factor analysis using these observed variables extracted two latent factor variables which are presented in Table 1 under factor loadings and identified as F_{1} and F_{2 }(Eigenvalue Total ≥ 1). These values represent the correlation (the standardized values) between the observed indicators as they covary between each other in relation to the factor loading. F_{1} measures the “political” factor and F_{2} measures the “social” factor. As an example, Table 1 reflects that the correlation between indicator 1 and F_{2} is .997. This means that concern about extremism loaded high in correlation to the Social Factor for attitude toward use of violence. 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 F_{1} and F_{2} together. Therefore when examining the communality value for indicator one it may be concluded that F_{1} and F_{2} together explain about 99% of variance in how concerned a respondent was about the possible rise of Islamic extremism in the U.S. in 2007.
____________________________
Table 1 About Here
____________________________
Turning to Table 1 F_{2} reports a low factor loading for the indicators presented in rows 2, 3 and 4. However, indicator 2 (do you feel suicide bombing and other forms of violence are justified to defend Islam from its enemies) and indicator 3 (do you have a favorable or unfavorable opinion of al Qaeda) each are highly correlated with F_{1}. When it comes to feelings about suicide bombing and violence F_{1} and F_{2} explain about 52% of variance in feelings. Likewise these two factors explain about 65% of variance in how favorable or unfavorable a respondent felt about al Qaeda in 2007. F_{1} has a positive, moderate correlation to indicator 4 (do you think the U.S. made the right decision in using military force in Afghanistan).
Turning to the total variance explained the Eigenvalues reported both F_{1} and F_{2 }values over 1, however it is noteworthy that F_{2} is marginal (F_{1} = 1.462; F_{2} = 1.001). F_{1} explains about 37% of the covariation around these four indicators while F_{2 }explains about 25%. Together these two factors explain about 62% of the covariance among the four indicators.
Regression was then conducted in which each of the two factors were used as the dependent variable (DV) in a separate model. The predictor variables selected for regression were Age, an interval/ratio level variable; Sex, which was dummy coded Male = 1, Female = 0; Financial Situation, ordinal recoded to four categories where 1 is “excellent shape” and 4 is “poor shape”; Importance of Religion, ordinal recoded to four categories where 1 = “very important” and 4 = “not at all important”; and last Education Attained. This variable was recoded ordinal level with seven categories with 1 = least amount of school and 7 = greatest amount of school.
It is noted that since the majority of indicators were ordinal with 4 categories regression was first attempted using Ordinal Regression technique. Two models were generated using F_{1} as the dependent variable in one model, and using F_{2} as the dependent variable in the other model. The Test of Parallel Lines in each model indicated a highly significant value (p = .001) therefore calling for a rejection of the null hypothesis that the location parameters are the same across response categories. Accordingly the assumption of parallel lines was not met and Ordinal regression is not proper to use. The SPSS report is attached as an appendix. Hence, Ordinary Least Square regression (OLS) was performed twice using F1 as the dependent variable in one model 1, and using F2 as the dependent variable in Model 2. The findings are presented in Table 2 below.
____________________________
Table 2 About Here
____________________________
Table 2, model 1 presents the OLS regression estimates predicting attitudes disapproving use of violence, U.S. Muslims age 18 and over for 2007, N = 701. The table reports the unstandardized coefficient as B and the standardized coefficient as β, all standard errors are reported in parentheses. Examining the model statistics, the predictors together in Model 1 explain about 10% of variation in attitudes disapproving use of violence in U.S. Muslims age 18 and over in 2007. In Model 2, the predictor variables together explain only about 2% of variation in variation in attitudes disapproving violence. In model 1 Sex where male = 1 and female = 0, and Financial Situation were not significant in predicting F_{1 }(p > .05). Importance of Religion was significant although weak (B = .119 (.047), p ≤ .05). Holding all other variables constant, for each level of greater importance of religion attitude disapproving use violence increases by .119. Age and Education are both positive and highly significant (p ≤ .001). The unstandardized B coefficient for Age indicates that controlling for all other predictors, for each year increase in age the attitude disapproving use of violence increases .128 level. Likewise for each level increase in education attitudes disapproving use of violence increases by .153 level, holding all other predictors constant.
Model 2 where F_{2} was used as the dependent variable shows that “importance of religion” makes no difference in attitude toward violence. Likewise, age, financial situation, and education make no difference in attitude toward violence, none are significant at .05 level. However, sex makes a difference in attitude toward violence (B = .296, p ≤ .001). Holding all other predictors constant men on average have .296 less favorable attitude toward violence than women. Surprisingly, men on average men approve use of violence less than women.
In conclusion I find this interesting and intend to go back and reverse code the two responses in the indicator variable “AfghanWarDecision” where response 1 Right Decision = 2, and response 2 Wrong Decision = 1. While it appears that these are coded in the right direction, it will be interesting to see what change may occur in the reported component matrix value and rotated component matrix value for this indicator.
Table 1: Factor Loadings of Rotated Factors on Attitudes Disapproving Use Of Violence, U.S. Muslims Age 18 and Over, 2007. (N = 701)
Factor loadings 
Communality 

Variable  F_{1}
_{(Political)} 
F_{2}
_{(Social)} 
h_{j}^{2} 

.006  .997  .993 

.720  .066  .523 

.805  .006  .648 

.543  .056  .298 
Eigenvalues  1.462  1.001  
% of variance explained  36.558  25.023 
Notes: Extraction method: principal components analysis
Rotation method: Varimax with Keiser normalization
Table 2: OLS Regression Estimates Predicting Attitudes Disapproving Use Of Violence, U.S. Muslims Age 18 and Over, 2007. (N = 701)
Political Factor Scale Model 1 (F_{1}) 
Social Factor Scale 

Predictor  B  β  B  β 
Importance of religion  .119*
(.047) 
.093  .055
(.048) 
.043 
Age  .128***
(.039) 
.120  .009
(.040) 
.009 
Male  .122
(.074) 
.060  .296***
(.076) 
.148 
Financial Situation  .002
(.044) 
.002  .063
(.046) 
.055 
Education  .153***
(.025) 
.239  .021
(.026) 
.033 
Constant  1.425***
(.214) 
.206
(.220) 

Adjusted R^{2}  .101  .024  
N  701  701 
*p ≤ .05 **p ≤ .01 ***p ≤ .001
Note: Standard errors are in parentheses.
Source: Pew 2007 Muslim American Survey
When Citing: Rigsby, Malcolm L. 2010. “Attitudes Disapproving Use Of Violence, U.S. Muslims Age 18 and Over, 2007.” Retrieved ____ (https://socialworld1.wordpress.com/wpadmin/postnew.php).