Tear Tracking Trends
Elizabeth Oyler and Taylor Patterson
April 25th, 2024
Data Science and Literacy
Maguire - 5th period
Introduction
Over the course of 2023 and the beginning of 2024, Elizabeth Oyler used the TrackingMyTears app to record the details of every cry she had. Although she did not always record her cries right after the event, she recorded the data within a few days of the cry. Elizabeth is a high-schooler in the graduating class of 2024 at the University of Chicago Laboratory Schools in Chicago, IL. She is a straight, white, able bodied, 17 year old (as of April 2024), cis female, and upper middle class. She is a swimmer, works various jobs, is a dedicated student, has two parents, a 16 year old brother, and a small dog. Overall we aim to discover, first what are the main factors that impact her cries the most? Then with that information - How does her frequency and reason of crying compare to the average human and average teenage girl? To begin to examine this we created a timeline located below of major events in her life in 2023 and 2024 that triggered strong emotions and are relevant to her tears tracking:
The visualization above depicts a timeline of January 2023 to
April of 2024. The Key shows that the blue water at the bottom is
swimming, the purple marks are travel, the small red marks are her
period, the orange represents therapy, and the relationships are pink.
The exclamation marks represent major events. The tick marks on the
green timeline represent school. The significant patterns in these
trends are marked by her relationships as well as outside events. This
visualization suggests to us that the more events going on at once, the
more likely tears will occur. We expect that major relationship events
during school times are one of the most frequent tear inducing
events.
It is generally found that there are three different types of tears. In our study we included all types of tears as we defined a tear as not tearing up but any tear that leaves the eyes for any reason. Primary there are three types of tears: Basal, Irritant, and Emotional. Basal tears are ones that coat the eyes all day and blinking will help spread them evenly amongst the surface of our eyes. They are primarily present to protect our eyes and keep our debris. Irritant tears appear when an outside force irritates the eye and tears are needed to flush out the aggravator. Usually these tears occur when you peel onions, throw up, or something nests itself in our eyes. Lastly emotional tears hold true to the name and arise from strong emotions. The emotion does not just have to be sadness. These tears come from physical pain, attachment pain, anger, passion, or any other strong emotion.
In the U.S it is estimated that American Women cry from an average of 3.5-5.3 times a month while men cry on average 1.3-1.9 times a month. In comparison Elizabeth cried 146 times in 2023 which was 12.1 times per month, and 172 times by the end of March 2024 which would be 11.4 times per month. Overall the rate of cries in 2024 are around 8.6 times per month. From our research, we found that social media has been the primary modern cause of tear inducing emotion. Human beings are social creatures who value social connection which is proven to relieve stress and sadness. Due to this urge, social media has become a crucial part of people’s daily activities causing the worldwide use to be over 3.484 billion in 2019, and up to a 9% increase year-on-year. Due to this frequent use, mental health has become connected to social media use. Mental health is widely defined as the state of well-being in which people understand their abilities, solve everyday problems, work well, and make a significant contribution to the lives of their communities. Because social networking is a crucial element in protecting our mental health, the quantity and quality of social relationships, including those on social media, affect our mental health. The Displaced Behavior Theory may help explain why social media shows a connection with mental health. On the other hand, social theories found how social media use affects mental health by influencing how people view, maintain, and interact with their social network. Furthermore, social media has been proven to relate to negative symptoms of depression, anxiety, and stress. In part this is credited to social media’s broadcasting of stereotypes and “ideal lifestyles.”
Specifically for teens, it has been recorded that they spend 4.8 hours on social media each day which is a significant amount of consumption. The frequency of social media use especially for women has increased depression and anxiety. Girls spend nearly an hour more on social media than boys (5.3 vs. 4.4 hours, respectively). Women were found to be more addicted to social media than men and tend to place more value on interpersonal trust and bonding through social media.
To examine the patterns of her tears, we created a specialized data set that would account for both emotions and locations. Our variables included the number of times she cried, the year/month/day/time that she cried, where she cried, the emotions that felt, how long she cried for, and why. We specialized two variables in the data set in order to ease the complexity in measuring them. With the “where” category we ensured that there were only a certain number of locations by summarizing them under broader terms. These included “Home, My Room, Alex’s Room, Swim Practice, Restaurant, School, Work, Outside, Someone Else’s House, Airplane, Car, Therapy, Swim Meet, and Library”. Furthermore we standardized emotions which were “Anger, Anxiety, Disappointment, Disgust, Fear, Frustration, Grief, Guilt, Happiness, Joy, Loneliness, Overwhelmed, Sadness, Shame, Stress, Surprise, Tired, Vulnerability, and Other. ’’ and we filled these out in a binary order where if she felt the emotion a number 1 would be placed under it, and if she did not a 0 would be placed under it. Additionally we categorized the length of crying in these following time frames: 1 tear, under 5, 5-19 minutes, 20-45 minutes, and 45+ minutes. Finally the “why” category is similar to the “where” in which we ensured a certain number of reasons including “Family, TV Show, Existentialism, BV, Writing, School, Hard Conversation, Losing Control, Body Image, Goodbyes, Reflection, Break-ups, Reading messages, Physical Discomfort, College, Sensory Overload, Driving, Therapy, and Tiktok Poetry.”
We have identified our null hypothesis based off of the most recent gender study(1980s) on men and women’s average crying. Our null hypothesis is 5.3 cries per month according to William H. Frey, PhD. We believe that this is not accurate because this estimate is not taking modern factors into account, so our alternative hypothesis states that null is not equal to 5.3 cries per month. H0 : mu = 5.3 cries per month, HA : mu ≠ 5.3 cries per month. Our P value was 11 percent which means that our test fails to reject the null hypothesis which indicates that 5.3 cries per minute is intact.
Visualizations and Data Analysis
For our first visualization above we created a graph that depicts how many times per each hour of the day that she cried. We wanted to understand which hours she cries the most, as well as could the hours indicate a preference of location. We found that the most frequent hours were 11am and 4pm, and the least frequent were 2am, 4am, and 6am. The least common hours are due to her sleeping schedule, and frequently she is asleep at these hours. A trend that we identified was that as the day progresses, the more likely crying will occur. The later hours in the day have more frequency in cries than the earlier ones. 11am and 4pm are significant hours in the day due to the presence of school at 11am and the end of school at 4pm. In later visualizations we identify that school and stressful situations are both prevalent reasons for why her crying occurs, and these findings align with the times of the day that she most frequently cries.
Our second visualization is an extension of our first where now that we understood the most common times, we took a look at where she most often cried to see if they would align. Furthermore we wanted to understand which locations would provoke the most tears. The valuation above depicts how many times she has cried at various locations. The most frequent locations are home, her room, school, and swim practice. 2 out of 4 of the common locations are ones that she appears in often between the hours of 11am and 4pm. On top of that both swim practice and school are stressful locations which were proven to be within the top reasons/emotions that she cried. Furthermore the locations she cried the least are the library, swim meet, work, and airplane. The scarcity locations indicate that she cries more often where she feels the most comfortable to show emotion or where she is overwhelmed past the point of concealment.
Next we wanted to explore in depth the reasons behind what provokes the most tears through the days. This visualization above depicts how many times she’s cried based on reason. We found that the most prominent reasons are hard conversation, reflection, school, family, stressful events, and losing control. The least common reasons are rasing, therapy, tiktok poetry, and driving. As we analyzed the various reasons we concluded that she cries the most often when she feels stressed or overwhelmed which she often feels during the previously mentioned prominent reasons she cries.
Finally we created a visualization to understand what emotion she was feeling the most often. The visualization above depicts how many times she felt each emotion while crying. The most prominent emotions were overwhelmed, tired, vulnerability, stress, frustration, and anxiety. The least common were joy, grief, and disgust. This visualization is an interesting one because the most prominent ones are the ones she felt most often that trigger the tears, yet some of the least felt ones could still be underlying emotion. However, the emotions that she feels the most align with the earlier findings with location, reason, and time.
Modeling
Lastly we created a linear model for the relationship between hour and month in 2023. The X axis represents months in numerical order. The Y axis represents time where 0 starts at midnight.
First we created an initial visualization to plot the relationship between the hour of day that Elizabeth cried given the month that it was. From a first glance it looks like the amount of cries per given time stays consistent throughout the year with slight variation in certain months. The months that had the most cries per house were January, June, August, and October. The times she cried the least were February and September. Finally the months with the most variation in the time that she cried were January and June. Furthermore we saw a slight downward trend and predicted that downward trend is what the linear model would depict.
When asked for the summary data for the linear model for the visualization above, below is what we received.
Call: lm(formula = tears2\(hour ~ tears2\)Month)
Residuals: Min 1Q Median 3Q Max -17.3852 -3.8048 0.7755 4.2256 9.0581
Coefficients: Estimate Std. Error t value Pr(>|t|)
(Intercept) 17.6982 0.8833 20.035 <2e-16 ** tears2$Month -0.3130
0.1330 -2.354 0.0199
— Signif. codes: 0 ‘’ 0.001 ‘’ 0.01 ‘’ 0.05
‘.’ 0.1 ‘ ’ 1 Residual standard error: 5.334 on 144 degrees of freedom
Multiple R-squared: 0.03705, Adjusted R-squared: 0.03036 F-statistic:
5.541 on 1 and 144 DF, p-value: 0.01993
Below we can see the linear model overlaid with the original
visualization. 
In making the linear model above, the coefficients for the linear y=mx+b equation are B=y-int=17.6982 and the slope=m=-0.3130. This indicates that there is a slight negative linear association of hour of day decreasing as the months go on with the beginning of the year starting with cries around 5 or 6 pm.
Below is the Residual plot for the above visualization. 
When plotted, the residual plot results in the above visualization. The plot is mostly a random scatter with a slight positive linear association located within the middle of the graph. Furthermore there are more high values than low values on the graph. These observations could suggest that there may be a model that could be better fit.
As both the summary and the visualization show, the furthermost point below the linear model is around -17.4 or midnight hours below the predicted hour and the furthermost point above is around 9 hours above. While the times seem further apart, they are closer than they seem. These are the outliers where she either much earlier in the day or much later in the day than the linear model predicts. If you split up the residuals into quartiles, the middle 50% of the data lies between -3.8 and 4.2 hours of the linear model with the median being around 0.8 hours above the model. The residual standard error is around 5.334 on 144 degrees of freedom with the T-statistic being 5.541. If the T statistic is at least 3.95, the null hypothesis can be rejected, and with a value of 5.541 we can reject the null hypothesis that there is no association between time of day and time of year of cries. Furthermore The P value is within the acceptable 5 percent range at 0.01993 which means this linear model can be accepted as statistically significant.
Conclusions
Overall circling back to our question of what are the main factors that impact Elizabeths cries the most? Along with, How does her frequency and reason of crying compare to the average human and teenage girl?: we found that Elizabeth cries the most at 11am and 4pm. She cries the most when she feels overwhelmed, tired, vulnerable, anxious, and stressed which aligns with the most frequent locations of school, and swim practice, and the vulnerability aligns with home and her room. The modeling example found that our P value of 0.01993 was within the acceptable 5 percent range which allowed our tests to be accepted as significantly significant. Because the P is in the acceptable 5 percent we reject the null hypothesis for our alternative hypothesis that there is a relationship between hour of day and month of cry. For future reference, we understand that Elizabeth as one example is not indicative of the whole modern teenage female population. If completing a holistic assessment we would need to repeat the same process with a variety of teenage women, yet we would block for different identifiers. Furthermore we would want to include more data on social media usage in addition to factors that often affect women like menstrual cycles or hormonal factors.
Code dump
set up
library(tidyverse)
library(modelr)
library(lubridate)
changing the data
tears<- tears|> separate(Time, into = c(“date”, “time”), sep = ” “)
tears<- tears|> separate(time, into = c(“hour”, “minute”), sep = “:”) View(tears)
tears2<- tears|> filter(Year==“2023”) View(tears2)
emotion totals
emotions<- tears |> select(Anger, Anxiety, Disappointment, Disgust, Fear, Frustration, Grief, Guilt, Happiness, Joy, Loneliness, Overwhelmed, Sadness, Shame, Stress, Surprise, Tired, Vulnerability)
emotions<- tears |> select(Anger:Vulnerability)
emotions2 <-emotions|> pivot_longer(cols = everything(), names_to = “Emotion”, values_to = “val”) %>% group_by(Emotion) %>% summarise(Total = sum(val))
Visualizations
ggplot(data=tears)+ geom_bar(aes(x=hour))
ggplot(data=tears)+ geom_bar(aes(x=Where))+ theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust = 1))
ggplot(data=tears)+ geom_bar(aes(x=Why))+ theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust = 1))
ggplot(data=tears)+ geom_bar(aes(x=Lengthmin))
ggplot(data=emotions2)+ geom_col(aes(x=Emotion, y=Total)) + theme(axis.text.x = element_text(angle = 45, vjust = 1, hjust = 1))
t test - the rest on calculator
sd(tears$Cry)
csv file export
write.csv(tears, file = “tears.csv”)
modeling attempt 1- had issues so not used:
Create an initial visualization
tears2<- tears|> filter(Year==“2023”)
ggplot(data=tears2)+ geom_point(mapping=aes(x= Month, y=hour, color=Lengthmin))
Fit models
m1 <- lm(hour ~ Month+Lengthmin, data = tears2) m2 <- lm(hour~Month*Lengthmin, data = tears2)
Gather predictions
grid <- tears2|> data_grid(Month, Lengthmin)|> gather_predictions(m1, m2)
grid<- grid |> add_predictions(m1)
tears2 <- tears2|> add_residuals(tears2)
grid <- tears2 |> data_grid(Month,Lengthmin) |> gather_predictions(m1, m2)
Plot predictions and compare between models by using facet_wrap (~model)
ggplot(tears2, aes(Month, hour, color=Lengthmin))+ geom_point()+ geom_line(data=grid, aes(y = pred))+ facet_wrap(~model)
Gather residuals
tears2 <- tears2 |> gather_residuals(m1, m2)
Plot residuals and facet on model ~ categorical variable
ggplot(data=tears2, aes(Month, resid, color = Lengthmin))+ geom_point() + facet_grid(model ~ Lengthmin)
ggplot(tears2, aes(tears2\(Month, resid, color=tears2\)Lengthmin))+ geom_point()+ facet_grid(model ~ tears2$Lengthmin)
ggplot(tears2, aes(Month, resid))+ geom_point()
modeling attempt 2- not as pretty but runs
scatterplot of raw data
plot(tears2\(Month, tears2\)hour)
computer output of lsrl
summary(lm(tears2\(hour~tears2\)Month))
adding model to scatterplot
abline(lm(tears2\(hour~tears2\)Month))
residuals
resid(lm(tears2\(hour~tears2\)Month))
residual plot
plot(tears2\(Month, resid(lm(tears2\)hour~tears2$Month)))
logarithmic model
plot(log(tears2\(Month), tears2\)hour)
Works Cited
Newhouse, L. (2001, March 1). Is crying good for you? Harvard Health Publishing; Harvard Health Blog. https://www.health.harvard.edu/blog/is-crying-good-for-you-2021030122020
O’Neill, A. (2001, May 24). Adolescents’ Understanding of What Causes Emotional Distress: A Qualitative Exploration in a Non-clinical Sample Using Ideal-Type Analysis. National Library of Medicine. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8181134/
Rothwell, J. (2023, October 3). OCTOBER 13, 2023 Teens Spend Average of 4.8 Hours on Social Media Per Day. Gallup. https://news.gallup.com/poll/512576/teens-spend-average-hours-social-media-per-day.aspx#:~:text=Teens%20Spend%20Average%20of%204.8%20Hours%20on%20Social%20Media%20Per%20Day