Social Factors for Student Alcohol Consumption
Cha Lor + Faith Opakunle
1/27/2020
Cleaning Vars
student.mat <- read.csv("~/Mscs 150 I20/Project/Cha-Faith_Project_Folder/student-mat.csv")
student.mat$avggrade = (student.mat$G1 + student.mat$G2 + student.mat$G3) / 3Cha’s pipes for GRADES BY SEX AND AGE
PIPING
CLEAN DATA
gradesavg <- student.mat %>%
group_by(sex, age) %>%
summarize(N=n(), mean=mean(avggrade)) %>%
ungroup()
gradesavg= gradesavg[-c(13,14),]grades_sex <- gradesavg %>%
group_by(sex) %>%
summarize(N=n(), avg=mean(mean))%>%
ungroup()Cha’s Visuals for GRADES BY SEX
ggplot(gradesavg, aes(x=factor(age), y=mean, fill=factor(sex), group=factor(sex)))+
geom_col(position="dodge", color="black")+
labs(x= "Age", y = "Average Grade", title = "Average Grade by Age and Sex")
ggplot(grades_sex, aes(x=sex, y=avg, fill=sex))+
geom_col()+
labs(x = "Sex", y = "Average Grade", title = "Average Grade by Sex")
#No difference in grades between both sexesThrough these two simples graphs, I wanted to introduce our topic of the average grade obtained by a student. I simply wanted to know if there was a differene between grades based on sex. And, like I expected, there’s no difference between grades by sex. But now we can ask… What factors are most significant when we look at a student’s grade? With this we explore many factors and their relationship with the average grade of a student.
Faith’s visual for average grades based on failures and age
p = ggplot(student.mat, aes(x=avggrade)) + geom_density(aes(fill=factor(failures)), alpha=0.8) + facet_wrap(~age) +
labs(title="Average Grades: Amount of Failed Classes by Age", x="Average Grades", y="Count", caption= "Source: student.mat")
p## Warning: Groups with fewer than two data points have been dropped.
## Warning: Groups with fewer than two data points have been dropped.
## Warning: Groups with fewer than two data points have been dropped.
## Warning: Groups with fewer than two data points have been dropped.
#This density plot showcases the average grades by the class and compares them to the number of times they failed a class and then facet wrapped by age. As we see in ages 21 and 22, there wasn’t enough data to be showcased and so those points were dropped. The most significant showing would be for age 17 where people who had failed at least 2 classes had an average grade of about 5 and the count was basically to the maximum of about 0.6. What is intresting is that the point for failing 2 classes and for the graph of age 16 of a point failing 3 classes with just a count of 0.2, has prominent “2 humpback points” but not as much variation in terms of average grades.The age group of 20 only had one point of students failing 2 classes but not going over the count of 0.2. This visual can tell that there’s not really a connection between the number of classes a student has failed and their average grades although there seems to be some outliers that changes the graph in some.
Cha’s Model For GRADES BASED ON ALCOHOL
alcmodel <- lm(avggrade ~ Dalc + Walc + studytime + goout + absences + freetime, student.mat )
summary(alcmodel)##
## Call:
## lm(formula = avggrade ~ Dalc + Walc + studytime + goout + absences +
## freetime, data = student.mat)
##
## Residuals:
## Min 1Q Median 3Q Max
## -9.2177 -2.2942 -0.0677 2.4414 8.6864
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 10.310958 0.949040 10.865 < 2e-16 ***
## Dalc -0.150536 0.273953 -0.549 0.58298
## Walc 0.081020 0.203005 0.399 0.69004
## studytime 0.597047 0.228279 2.615 0.00926 **
## goout -0.565659 0.188238 -3.005 0.00283 **
## absences 0.006819 0.023248 0.293 0.76943
## freetime 0.281276 0.196304 1.433 0.15270
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 3.641 on 388 degrees of freedom
## Multiple R-squared: 0.0449, Adjusted R-squared: 0.03013
## F-statistic: 3.04 on 6 and 388 DF, p-value: 0.006441alcstats= tidy(alcmodel) %>%
round_df()
alcconf <- tidy(alcmodel, conf.int = TRUE)
alcconf = subset(alcconf, term %nin% "(Intercept)") %>%
round_df()alcanova <- anova(alcmodel)
alcfitted <- fitted(alcmodel)
alcresid <- resid(alcmodel)
alcaov <- aov(alcmodel, student.mat)
alcaovtidy <- tidy(alcaov, conf.int = TRUE) %>%
round_df()I used an ANOVA model to represent the significance each variable had on a student’s grade. An ANOVA model is similar to a linear model, but is instead used for categorical values. I am then given the numbers that represent a category’s F-value, and their P-value. The F value is similar to a standard deviation, and it tells me how is value from the mean or average observation. The further it is away from the mean observation, the less likely it is bound to occur. The P value is similar to the F value, in which it means how likely will this event occur. For this specific case, I wanted to analyze the effect alcohol had on students’ grades. I did so by creating an anova model based on variables that matched the theme of alcohol consumption. If you look at the studytime and go-out variable, we can see that because the P value is very low, which tells us that those variables are very significant in determining a student’s grade.
summary(alcaov)## Df Sum Sq Mean Sq F value Pr(>F)
## Dalc 1 28 28.31 2.136 0.14471
## Walc 1 16 15.64 1.180 0.27800
## studytime 1 72 71.53 5.396 0.02069 *
## goout 1 99 98.73 7.449 0.00664 **
## absences 1 0 0.37 0.028 0.86760
## freetime 1 27 27.21 2.053 0.15270
## Residuals 388 5143 13.25
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1#values with asterisks suggest statistical significanceggplot(alcaovtidy, aes(x=term, y=p.value))+
geom_point() + coord_flip()## Warning: Removed 1 rows containing missing values (geom_point).
It didn’t make sense to plot my anova model P-values, just because it doesn’t mean much by itself if the audience doesn’t understand a p value. So instead I plotted the linear model with the confidence intervals instead, which essentially tells the same data. Going out and studying has major impacts on an indiviual’s grade, but having free time and alcohol consumption wasn’t a big factor. If you look at the P-values from the ANOVA model, this is also supported. This is a better model to represent the significance of variables, because the further away a term is from the zero line, the more significant it is in determining a student’s grade.
ggplot(alcconf, aes(x=reorder(term, estimate) , y=estimate,
ymin=conf.low,
ymax=conf.high))+
geom_pointrange()+
coord_flip() +
geom_hline(yintercept = 0, size = 1.4, color = "black", alpha=.5)
Cha’s Visual for GRADE BASED ON GOING OUT AND ALCOHOL
gooutmod <- student.mat %>%
group_by(sex, age, goout, studytime, Dalc, Walc, avggrade) %>%
summarize(N=n())%>%
summarize(averagegrade = mean(avggrade)) %>%
ungroup()
gooutmod <- mutate(gooutmod,
studytime = factor(studytime, labels=c("<2 hours", "2 to 5 hours", "5 to 10 hours", ">10 hours"), ordered = T),
Dalc = factor(Dalc, labels=c("very low", "low", "moderate", "high", "very high"), ordered = T),
Walc = factor(Walc, labels=c("very low", "low", "moderate", "high", "very high"), ordered = T))This graph is can be used to visualize our grade based by alcohol consumption model. If we look at the weekly consumption of alcohol, we see that the points don’t really follow a specific pattern; they’re very spread out and isn’t really clumped up anywhere. When we look at the amount of studying, we can see that the more a student studies, the higher their grades are. We can see this because if we look at only the blue trend, it’s higher on the y axis than any other trend. When we look at the statistically significant variables, study time and going out, we see that there is a very subtle pattern between the two variables. Those who study more, have a trend at a higher y intercept, but as the rate of going out increases, all trends decrease; in other words, the more a student goes out, the lower their grade. Even though its very subtle, those who study more, the red and blue trends, are shown to have a higher grade than those who study less, the yellow and green lines. By looking at this visual, we can rephrase our initial hypothesis of thinking that alcohol consumption will significantly impact a student’s grade. We can instead say that the grade a student obtains isn’t based on alcohol consumption, but is more influenced by the amount a student studies and the amount a student goes out. this hypothesis is supported by our summary of variables, which tells us that going out and study time are statistically significant, whereas alcohol consumption has no significant effect. when we look closely at this visual, we see that the summary of variables supports our hypothesis as well.
alccolors <- c("#44cf4d", "#cfcd00","#cf4449", "#444dcf")ggplot(gooutmod, aes(x= goout, y=averagegrade, color=factor(studytime), group=studytime))+
geom_jitter(alpha=.5)+
labs(x= "Going Out\nScale from 1(Low) to 5(High)",
y= "Average Grade",
title = "Average Grade based on amount of Going Out",
subtitle = "Size is based on weekly alcohol consumption",
color= "Study Time",
size= "Weekly Alcohol\n Consumption Scale\n from 1(Low) to 10(High)")+
geom_smooth(method="lm", se= FALSE, size=1.5)+
scale_color_manual(values = alccolors)
ggplot(gooutmod, aes(x=factor(goout), y=averagegrade, fill=factor(studytime)))+
geom_boxplot()+
labs(x= "Going Out\nScale from 1(Low) to 5(High)",
y= "Average Grade",
title = "Average Grade based on amount of Going Out",
color= "Study Time",
size= "Weekly Alcohol\n Consumption Scale\n from 1(Low) to 10(High)")+
scale_fill_manual(values = alccolors)
BUT! When we look at this glance function, we’re given an R Squared value, and this value tells us how much of our statistically significant findings apply to our whole dataset. And in this case, we found that what we thought was initally statistically significant, going out and studying, it wasn’t as meaningful as we presumed. That’s because our data that showed it was significant, only applies to 4.5% of the dataset.
glance(alcaov)## # A tibble: 1 x 11
## r.squared adj.r.squared sigma statistic p.value df logLik AIC BIC
## <dbl> <dbl> <dbl> <dbl> <dbl> <int> <dbl> <dbl> <dbl>
## 1 0.0449 0.0301 3.64 3.04 0.00644 7 -1067. 2151. 2183.
## # … with 2 more variables: deviance <dbl>, df.residual <int>Faith’s visual for average grades based on famsup & alcohol consumption
alcsch <- c("#0ed9f0","#efbaf7", "#79fca9", "#f7c886", "#f2aba7" )Mutating Workday and Weekend Alcohol Consumption
student.mat <- mutate(student.mat,
Dalc = factor(Dalc, labels=c("very low", "low","medium", "high", "very high"), ordered=T),
Walc = factor(Dalc, labels=c("very low", "low","medium", "high", "very high"), ordered=T))p = ggplot(student.mat, aes(x=famsup, y=avggrade, fill=factor(Walc))) + geom_boxplot() +
labs(title="Average Grades: Family Support by Alcohol Consumption", x="Family Support", y="Average Grades", fill= "Alcohol Consumption During the Week", caption= "Source: student.mat") + scale_fill_manual(values=alcsch)
p
#I wanted to have the scale for weekday and workday alchohol be more coherent and make more sense for the graph instead of just numbers. So I mutated those 2 variables into a word scale of “very low” to “very high”. Based on the boxplot I did with just workday alcohol consumption, we can see that there is a variation of having medium alcohol consumption during the week if they have their family support and that only gives them the average in their grades. It’s a very different visual when we see for students who did not have any family support but still had a medium alchohol consumption in being that there are some varitation but not as much. This visual can tell us that whether or not a student has the support of their family or not, their intake of alcohol during the week is not that much difference for the average grades with the exception of some.
Cha’s model for GRADE BASED ON FAMILY RELATIONSHIP
ANOVA Model
In this model, I wanted to find if there was any relationship between a student’s average grade and their family relationship. So I made an ANOVA model to compare qualitative variables with a student’s average grade. This is the same process as I went through earlier with the ANOVA model for the theme of alcohol consumption.
famanova = aov(avggrade~ Pstatus + Medu + Fedu + Mjob + Fjob + famsup + famrel + higher, student.mat)
famfitted = fitted(famanova)
famresid= resid(famanova)
famtidy =tidy(famanova, conf.int= TRUE) %>% round_df()
view(famtidy)From this model, we found that the variables, mother’s education, and family support are statistically significant. Again, it shows us the likelihood of this event happening, and because the value is so low, as in, it’s very unlikely to happen, there must be a reason why it does occur, meaning its statistically significant.
ggplot(famtidy, aes(x=p.value, y=term))+
geom_point()## Warning: Removed 1 rows containing missing values (geom_point).
Again, it doesn’t really make sense to plot a p value, instead we’ll create a linear model to view the estimate and its confidence level.
Linear Model
famlm = lm(avggrade~ Pstatus + Medu + Fedu + Mjob + Fjob + famsup + famrel + higher, student.mat)
famstats= tidy(famlm) %>%
round_df()
famconf <- tidy(famlm, conf.int = TRUE)
famconf = subset(famconf, term %nin% "(Intercept)") %>% round_df() # Removes Grades intercept
view(famconf)Visuals
ggplot(famconf, aes(x=term, y=estimate, ymin=conf.low, ymax=conf.high))+
geom_pointrange() + coord_flip() +
geom_hline(yintercept = 0, size=1.3, color = "black", alpha=.5)
From this model, we can see that a mother’s education and having desire to pursue education positively correlates to a student’s grade, and ironically, having family support negatively correlates a student’s grade.
Ploting the estimate of the family model
ggplot(famconf, aes(x=reorder(term, estimate), y=estimate, ymin=conf.low, ymax=conf.high))+
geom_pointrange()+
coord_flip()+
geom_hline(yintercept = 0, size = .8, color = "black", alpha=.5)
Piping to get a mean grade
fammod <- student.mat %>%
group_by(Pstatus,Medu,Fedu,Mjob,Fjob,famsup,famrel,higher, avggrade) %>%
summarize(N=n())%>%
summarize(avggrade = mean(avggrade)) %>%
ungroup()fammod <- mutate(student.mat,
Fedu = factor(Fedu, labels=c("none", "primary education","5th to 9th grade", "secondary education", "higher education"), ordered=T),
Medu = factor(Medu, labels=c("none", "primary education","5th to 9th grade", "secondary education", "higher education"), ordered=T),
Mjob = factor(Mjob, labels=c("teacher", "health care related","civil services", "at home", "other"), ordered=T),
Fjob = factor(Fjob, labels=c("teacher", "health care related","civil services", "at home", "other"), ordered=T),
studytime = factor(studytime, labels=c("<2 hours", "2 to 5 hours", "5 to 10 hours", ">10 hours"), ordered = T))Cha’s Visuals for grade by famsup and want to pursue higher edu
From this visual, we can see those who does have their family’s support, ironically have a lower average grade. I would assume that their would be positive correlation, but the plot of the confidence interval and this comparative boxplot tells us otherwise.
ggplot(fammod, aes(x= Medu, y=avggrade, fill=famsup))+
geom_boxplot(alpha=.5, position="dodge", color="black")+
labs(y= "Average Grade",
x= "Mother's Education",
fill = "Family Support",
title= "Average Grade based on Mother's Education",
subtitle = "Separated by Family Support")+
theme(axis.text.x = element_text(angle = -10))
From this visual, we can see that those who want to pursue higher education have greater average grades than those who don’t. This matches with our thinking, we would think that if a student really likes school and wants to pursue higher education, they would work hard and do well, which our visual does support.
ggplot(fammod, aes(x= Medu, y=avggrade, fill=higher))+
geom_boxplot(alpha=.5, position="dodge", color="black")+
labs(y= "Average Grade",
x= "Mother's Education",
fill = "Want to\nPursue Higher\nEdu",
title= "Average Grade based on Mother's Education",
subtitle = "Separated by Desire to Pursue Higher Education")+
theme(axis.text.x = element_text(angle = -10))
glance(famanova)## # A tibble: 1 x 11
## r.squared adj.r.squared sigma statistic p.value df logLik AIC BIC
## <dbl> <dbl> <dbl> <dbl> <dbl> <int> <dbl> <dbl> <dbl>
## 1 0.114 0.0810 3.54 3.48 2.26e-5 15 -1053. 2137. 2201.
## # … with 2 more variables: deviance <dbl>, df.residual <int>But then again, when we look at this glance function, we’re given an R Squared value. And in this case, we found that what we thought was initally statistically significant, the education of a student’s mother, the student wanting to pursue higher education, and whether or not their family supports their education or not, it wasn’t as meaningful as we presumed. That’s because our data that showed it was significant, only applies to 11.5% of our dataset. This number is much more significant than the previous number we were given, but still, it doesn’t not contain the majority of the data.
Faith’s visual for average grades by reason and school sup
p = ggplot(student.mat, aes(x=reason, y=avggrade, fill=schoolsup)) + geom_boxplot() +
labs(title="Average Grades: Reason for School Choice by School Support", x="Reason for School", y="Average Grades", subtitle = "Fill color is based on school support", caption= "Source: student.mat")
p
#This visual is looking at students’ reasons for picking the school they did for their average grades and whether or not they had the support of their school. I used a boxplot to see the comparison of whether or not they had school support. For the reason of reputation and other, they had a distinct variation if they didn’t have school support and the average grades for it was above 10 and the two biggest boxplot. On the contrast, all the reasons that students had for picking their school had the lowest average grades, even if they did have school support. I find it very ironic because we would believe that students that have the support of their school would do better grade-wise but it’s not the case for this graph. I felt this kind of graph was necessary to actually see these variables between one another because we were expecting the opposite results.