You are on page 1of 19

By Aynur Aytekin

Introduction

 This project is to analyze the results of parent satisfaction surveys for Concept
Schools.
 Concept Schools is a charter school management company in the Mid-West area
with 30 charter schools serving more than 13000 students in 7 different states.
 Data set used in the project is the results of the parent satisfaction surveys from
2014,2015, and 2016. Total of 12813 people were participated in these surveys.
 The parents were asked 41 survey questions to evaluate their satisfaction on
school’s academic program, discipline, school culture and atmosphere, school
safety, school staff and administrators.
Demographics of the Participants:

Gender

Cumulative
Frequency Percent Valid Percent Percent
Valid

Male 3312 25.8 25.8 25.8

Female 9501 74.2 74.2 100.0

Total 12813 100.0 100.0


Ethnicity

Frequency Percent Valid Percent Cumulative Percent


Valid
African American 6766 52.8 52.8 52.8

White /Caucasian 2908 22.7 22.7 75.5

Hispanic 1902 14.8 14.8 90.3

Asian/Pacific Islander 280 2.2 2.2 92.5

Native American 69 .5 .5 93.1

Other 888 6.9 6.9 100.0

Total 12813 100.0 100.0


Date of the Survey Administered

Cumulative
Frequency Percent Valid Percent Percent
Valid

2014 3701 28.9 28.9 28.9

2015 4934 38.5 38.5 67.4

2016 4178 32.6 32.6 100.0

Total 12813 100.0 100.0


Variables
Categorical/Nominal : (Specified as factor)
Gender 1-Male
2-Female

Categorical/Nominal : (Specified as factor)


Independent Variables

1-African American
2-White/Caucasian
Ethnicity 3-Hispanic
4-Asian/Pacific Islander
5-Native American
6-Other
Date of the Survey
Quantitative/Integer : 2014, 2015, 2016

Quantitative/Numeric: Composite Score of 5 Survey


Items
5-”Strongly Agree”
Parents’ Academic Satisfaction from the School 4-”Agree”
3-”Not Sure/Neutral”
2-”Disagree”
1-”Strongly Disagree”
Dependent Variables

Categorical/Nominal : (Specified as factor)


Home Internet Access of the Parents 1-”Yes”
2-”No”

Categorical/Ordinal: (Specified as Ordinal Factor)


5-”Strongly Agree”
4-”Agree”
Parents’ Overall Satisfaction from School
3-”Not Sure/Neutral”
2-”Disagree”
1-”Strongly Disagree”
Research Questions:
Research Questions:
1. Do gender, ethnicity, and the date of the survey
significantly predict whether the parents have
home internet access or not?

2. Do gender, ethnicity, the date of the survey, and


the academic satisfaction of the parents from
school significantly predict the overall parent
satisfaction?
Predicting whether an individual has internet
access at home from ethnicity, gender and years.
Binary Logistic Regression Model
> logreg1<-glm(ParSRV$internetR~ParSRV$raceR+ParSRV$genderR+ParSRV$startdate,family
=binomial)
> summary(logreg1)

Call:
glm(formula = ParSRV$internetR ~ ParSRV$raceR + ParSRV$genderR +
ParSRV$startdate, family = binomial)

Deviance Residuals:
Min 1Q Median 3Q Max
-2.4954 0.4662 0.5463 0.5766 0.7312

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -137.41377 67.06116 -2.049 0.040454 *
ParSRV$raceRWhite/Caucasian 0.45423 0.07005 6.484 8.92e-11 ***
ParSRV$raceRHispanic 0.06234 0.07380 0.845 0.398257
ParSRV$raceRAsian/Pasific Islander 1.28881 0.28608 4.505 6.63e-06 ***
ParSRV$raceRNative American -0.45861 0.29409 -1.559 0.118892
ParSRV$raceROther 0.37618 0.11290 3.332 0.000863 ***
ParSRV$genderRFemale -0.04758 0.06012 -0.791 0.428737
ParSRV$startdate 0.06907 0.03328 2.075 0.037953 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

Null deviance: 10099 on 12812 degrees of freedom


Residual deviance: 10015 on 12805 degrees of freedom
Results:

• The p-value associated with the test statistics is very small. The p-value is
p<0.001(near zero).
• We reject the null hypothesis that the predictors are not associated with the home
internet access of the parents.
• The set of predictors do significantly predict the internet access of the parents.
(x²=84, p<.001)
• African American is the reference
category. White/Caucasian, Asian/Pacific
Islander and other ethnicity types are the
statistically significant predictors.
• The date of the survey is statistically
significant predictor.
• Hispanic, Native American is not
statistically significant predictors.
• Gender is not statistically significant
predictor.
Odds-Ratios
> exp(coef(logreg1))
(Intercept White/Caucasian Hispanic Asian/Pasific Islander
0.002098746 1.574957 1.064326 3.628484

Native American Other Female startdate


0.6321609 1.456709 0.9535344 1.071509

Compared to an African American, the odds that a White/Caucasian has home internet access are 1.575 times
larger.

Compared to an African American, the odds that a Hispanic has home internet access are 1.064 times larger, it
is almost the same value for African Americans and Hispanics.

Compared to an African American, the odds that a Native American has home internet access are 0.632 times
larger, or African Americans have home internet access 1.587 times more than Native Americans.

Compared to African American, the odds that other ethnicities with home internet access are 1.457 times
larger.

Compared to males, the odds that females with home internet access are 0.953 times larger.

Each unit increase in years is associated with a 1.071 times increase in the odds of having home internet access
for the parents.
Testing Statistical Significance of Ethnicity:
> logreg2<-glm(ParSRV$internetR~ParSRV$genderR+ParSRV$startdate,family=binomial)

> summary(logreg2) > summary(logreg1)

Call: Call:
glm(formula = ParSRV$internetR ~ ParSRV$genderR + ParSRV$startdate, glm(formula = ParSRV$internetR ~ ParSRV$raceR + ParSRV$genderR +
family = binomial) ParSRV$startdate, family = binomial)

Deviance Residuals: Deviance Residuals:


Min 1Q Median 3Q Max Min 1Q Median 3Q Max
-2.0476 0.5251 0.5271 0.5424 0.5560 -2.4954 0.4662 0.5463 0.5766 0.7312

Coefficients:
Coefficients:
Estimate Std. Error z value Pr(>|z|)
Estimate Std. Error z value Pr(>|z|)
(Intercept) -137.41377 67.06116 -2.049 0.040454 *
(Intercept) -121.79968 66.69954 -1.826 0.0678 .
ParSRV$raceRWhite/Caucasian 0.45423 0.07005 6.484 8.92e-11 ***
ParSRV$genderRFemale -0.05339 0.05987 -0.892 0.3726
ParSRV$raceRHispanic 0.06234 0.07380 0.845 0.398257
ParSRV$startdate 0.06139 0.03310 1.855 0.0636 . ParSRV$raceRAsian/Pasific Islander 1.28881 0.28608 4.505 6.63e-06 ***
--- ParSRV$raceRNative American -0.45861 0.29409 -1.559 0.118892
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 ParSRV$raceROther 0.37618 0.11290 3.332 0.000863 ***
ParSRV$genderRFemale -0.04758 0.06012 -0.791 0.428737
(Dispersion parameter for binomial family taken to be 1) ParSRV$startdate 0.06907 0.03328 2.075 0.037953 *
---
Null deviance: 10099 on 12812 degrees of freedom Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual deviance: 10094 on 12810 degrees of freedom
(Dispersion parameter for binomial family taken to be 1)
AIC: 10100

Null deviance: 10099 on 12812 degrees of freedom


Number of Fisher Scoring iterations: 4
Residual deviance: 10015 on 12805 degrees of freedom
Testing Statistical Significance of Ethnicity:

Δ Deviance=10094-10015=79
Δ df= 12810-12805=5

The p-value associated with this test statistics is smaller than 0.001. The
predictor “ethnicity” significantly predicts the home internet access for the
parents.(X² = 79, p < 0.001)
Do gender, race, the date of the survey administered, and the
academic satisfaction of the parents from school significantly
predict the overall satisfaction of the parents?
> ParSRV$AcExp<- rowMeans(cbind(ParSRV$AcExp1,ParSRV$AcExp3,ParSRV$AcExp4,ParSRV$AcExp5,P
arSRV$AcExp6),
+ na.rm=TRUE)

> boxplot(ParSRV$AcExp, +
main="Boxplot of Parents' Academic
Satisfaction \n from school", +
ylab="Composite Score", +
col="tomato", notch=TRUE)
Ordinal Regression Model for Predicting Parent Satisfaction from
Gender, Ethnicity, Year, and Academic Satisfaction from the School

> OrdinalReg1<-ordinal::clm(data=ParSRV,SchoolSatfR~genderR+raceR+AcExp+year1)
> summary(OrdinalReg1)
formula: SchoolSatfR ~ genderR + raceR + AcExp + year1
data: ParSRV

link threshold nobs logLik AIC niter max.grad cond.H


logit flexible 12813 -10305.95 20635.90 7(0) 3.35e-12 7.2e+03

Coefficients:
Estimate Std. Error z value Pr(>|z|)
genderRFemale -0.09046 0.04460 -2.028 0.04256 *
raceRWhite/Caucasian 0.26721 0.04907 5.446 5.15e-08 ***
raceRHispanic 0.39105 0.05843 6.692 2.20e-11 ***
raceRAsian/Pasific Islander 0.36951 0.13512 2.735 0.00624 **
raceRNative American 0.21458 0.25691 0.835 0.40358
raceROther 0.19274 0.07865 2.451 0.01426 *
AcExp 3.02382 0.03997 75.648 < 2e-16 ***
year1 -0.01511 0.02480 -0.609 0.54236
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
Estimate Std. Error z value
Strongly Disagree|Disagree 7.5155 0.1641 45.79
Disagree|Not Sure 9.3853 0.1663 56.42
Not Sure|Agree 10.1258 0.1689 59.95
Agree|Strongly Agree 13.5479 0.1898 71.37
Results of the Ordinal Regression Model

 Gender/Female (b=-0.089, p=0.044),


Race/White Caucasian (b=0.268, p=4.51e-08)
Race/Hispanic (b=0.390, p=2.37e-11)
Race/ Asian Pacific Islander (b=0.368, p=0.00644)
Race/Other (b=0.19, p=0.1363)
Academic Satisfaction (b=3.024, p<2e-16) are statistically significant predictors of parents’
overall school satisfaction.
 Race/Native American and the date of the survey are not statistically significant predictors for the
overall satisfaction of the parents from school.
To check overall model fit;
> OrdinalReg2<-ordinal::clm(data=ParSRVReg1, SchoolSatfR~genderR+raceR+year1)
> summary(OrdinalReg2)
formula: SchoolSatfR ~ genderR + raceR + year1
data: ParSRVReg1

link threshold nobs logLik AIC niter max.grad cond.H


logit flexible 12813 -14548.76 29119.53 7(0) 5.14e-08 1.1e+03

Coefficients:
Estimate Std. Error z value Pr(>|z|)
genderRFemale -0.083452 0.038333 -2.177 0.02948 *
raceRWhite/Caucasian 0.320434 0.042245 7.585 3.32e-14 ***
raceRHispanic 0.536129 0.049884 10.747 < 2e-16 ***
raceRAsian/Pasific Islander 0.326320 0.112873 2.891 0.00384 **
raceRNative American 0.003617 0.230347 0.016 0.98747
raceROther 0.029139 0.067900 0.429 0.66781
year1 -0.024146 0.021406 -1.128 0.25932
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
Estimate Std. Error z value
Strongly Disagree|Disagree -3.44065 0.07578 -45.402
Disagree|Not Sure -2.27063 0.06338 -35.823
Not Sure|Agree -1.80760 0.06106 -29.603 According to the anova results the
Agree|Strongly Agree 0.28006 0.05821 4.811
academic satisfaction of the
> anova(OrdinalReg2,OrdinalReg1)
Likelihood ratio tests of cumulative link models: parents from school is a statistically
formula: link: threshold: significant predictor of the overall
OrdinalReg2 SchoolSatfR ~ genderR + raceR + year1 logit flexible
OrdinalReg1 SchoolSatfR ~ genderR + raceR + AcExp + year1 logit flexible satisfaction of the parents from
no.par AIC logLik LR.stat df Pr(>Chisq) school. ( p<2.2e-16)
OrdinalReg2 11 29120 -14549
OrdinalReg1 12 20636 -10306 8485.6 1 < 2.2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
> ggplot(ParSRVsubset3, aes(x=SchoolSatf, y=AcExp, color=raceR))+ +
stat_summary(fun.y=mean,geom="line",aes(group=raceR))+ + labs(x="Parents' Overall Satisfaction from School",
y="Parents' Academic Satisfaction \n from School")+ + labs(title="Plot of Parents' Academic Satisfaction from School
\n by Parents' Overall Satisfaction from School \n and The Ethnicity")
Results

 According to the analysis of the survey ethnicity and the date


of the survey are the statistically significant predictors for the
home internet access for the parents.
 Ethnicity, gender and the parents’ academic satisfaction from
the school are statistically significant predictors of the parents’
overall satisfaction from the school.
 For the future further research about,
 If;
School Discipline,
Safety ,
School Culture,
School Staff and Administration,
and the other variables in the survey predict the satisfaction
of parents.

You might also like