Introduction

With the COVID-19 pandemic anniversary coming up in December, scientists across the world are racing against time to develop a vaccine for this deadly virus. The role that data analysis and visualization has played during this outbreak is far from imaginable. Data analysis not only made it possible for us to fight this virus but also identify and save communities. For my Carto Lab activity I analyzed the COVID risk according to the county population, and thought it would be timely and appropriate for my final project to be an extension of this lab.

I answered few of the major questions through my last lab, but it still wasn’t a complete analysis. Further research on the dataset raised a few interesting and previously unanswered questions, which I aim to answer through this project. The questions are as follows:

1. Can we run predictive analysis on the existing dataset to determine high risk counties and further perfect resource allocation?
2. What if the effect of wearing a mask on the number of cases?
3. How can we track the movement of the virus across US using spatial analysis?

Datasets and Softwares

I used the same datasets from my previous lab in addition to the following new ones.

Mask use by county: The dataset is originally sourced from New York Times
Latitude and Longitude of US counties

Like the previous project I used Carto for this project as well. But using only Carto for all my needs proved to be a little difficult this time, so I used OpenRefine for data cleaning and organization.

Process

Before starting my analysis, I cleaned and organized my data in OpenRefine. The next step was to import this data in Carto and draw my final spatial analysis.

Data Cleaning

Initially the data for the total number of COVID cases was organized in separate columns, and the latitude and longitude columns were in a different table all together. Since the dataset with county latitudes and longitudes was in a zip format I had to do a little bit of extra work in order to merge the tables and get the desired final dataset.

1. Import the county Latitude and Longitude dataset into Carto and download it as a CSV.
2. Import the datasets in OpenRefine and import the latitude and longitude rows into the original dataset using the following function.
   For adding Longitude row: cell.cross(“c_03mr20_1″,”fips”).cells[“lon”].value[0]
   For adding Latitude row: cell.cross(“c_03mr20_1″,”fips”).cells[“lat”].value[0]
3. Perform Transpose operation to interchange the rows and columns, and export the Final Dataset.

• 

• 

• 

Analysis

There are a lot of visualization out there to depict the COVID data, but their lack of highlighting the county population might result in inaccurate processing by the user. In order to tackle this problem I used the size variable to highlight the county population.

The first visualization I created aimed at highlighting the total number of COVID cases over time in every US county. I varied the point size of each county according to its population. Although I went back and forth while choosing the classification method for the point size, I finally settled for Jenks as it breaks the data into classes based on natural groupings inherent in the data. The final visualization has 5 data buckets and the size varies from 2 to 15. I also added a date widget which gives the user the flexibility to filter out and see only specific dates, in addition to this it also helps us analyze the propagation of the virus throughout the states.

The second visualization was aimed at conducting the predictive analysis on the US county COVID data, to determine which counties might act as the next hotspots. For running the Predictive Analysis I used the Predict Trends and Volatility (PTV) analysis tool built into Carto. PTV uses Spatial Markov Chains to calculate the trends and volatility. Since the system changes randomly, it is difficult to predict with certainty the next state of the Markov chain. However, the statistical probability of a systems future can be predicted.

The changes of states are called transitions, and the probabilities associated with various state-changes is called transition probabilities. The deviation of sum of these probabilities trending up (relative to the unit index of that probability) is given by trend_up, trending down by trend_down and overall trend (where sign signifies the direction negative for down and positive for up) by trend. Volatility is the degree of variation of the event series data over time, measured by the standard deviation of probabilities within the trends.

Through my third visualization is wanted to predict the number of cases based on the mask use data. So in order to do that I used the mask use dataset from New York Times, and based on this data I ran PTV analysis on my original dataset to see how the trend changes based on weather the county tends to use masks always or never. In order to facilitate user interaction with this visualization, I added the always and never widgets. These widgets let the user view weather a particular county is gonna trend up or down based on how they use the masks.

UX Study

In order to be able to cater to a larger audience, I wanted my visualization to be easy to interact with and comprehensible. I planned my user study with this idea in mind. I conducted my user study with a total of 2 participants, 1 out of the 2 participants had limited technological experience, but knew a lot about the COVID-19 pandemic. For this particular user study I figured that the think aloud method would work the best, as it would not only help me get a idea about how the user feels about the maps but also about the aesthetics like color and widget placements.

Rationale

While I was researching for initial project I came across a plethora of visualization for COVID, and each one of them had a distinct feature. But interestingly enough I did not come any visualization that uses the power of predictive analysis to show the user what the future trends may look like. The other important feature that most of the visualizations fail to depict was the county population and how it correlates to the number of cases.

I believe when combined together all the final three spatial visualizations do a better job at explaining the user the cause and effect of COVID-19. I also think that looking at the mask use predictive analysis visualization more people are going to be motivated to wear masks which would help us further combat this deadly virus.

I used a consistently same color scheme across all the three visualizations. In a general setting this would be considered a bad idea as the same color defines different things in different visualizations, but since my visualizations are standalone and can be viewed and interpreted individually I thought it would be a good idea to use the same color scheme across the spectrum. The colors are color blind safe, and also subtle enough to not distract the user from data.

Findings

Visualizations

Working together the maps deliver a much clearer picture of the COVID-19 pandemic. Based on the first visualization we can how it has spread across the US since the first case was detected. Looking at these trends we can also reflect on some of the steps that could have prevented it from getting this worse. The first look at the graph helps us conclude that the counties with higher population have higher rate of infection, but with exceptions in Maine and Vermont, where the rate of infection is quite low in spite of the large populations.

The mask use map also helps us give a clear picture of how important it is to wear a mask, as we can distinctively see that the counties with people “never” wearing a mask have a higher chance of being the next hotspot. The trend value from predictive analysis also shows how the number of new cases in most counties if going to decrease.

UX Study

While conducting the user study I was really glad that the users found it more helpful than any other COVID map they had previously used. One thing that all of the participants mentioned was the use of the point size to depict the county population, they thought it really helped them analyze the data much more effectively. Findings from the user study are noted below.

Findings from Participant 1

1. Some of the points are too small and light and can not be seen against the light background.
2. Clicking on the dates widget to filter a particular week, the following week is selected
3. Users don’t understand the meaning of the terminology like volatility, trend_up and trend_down.

Findings from Participant 2

1. It is harder to look for a particular county without the state borders
2. The user had difficulty scrolling to the states of Alaska and Hawaii

Recommendations Based on User Study

1. Use a darker base map for the main COVID map, as it highlights the lighter and smaller county points as well.
2. Add state borders to all of the maps to reduce the cognitive load on users end. As is it easier of find a particular county with reference to the state borders.
3. Adding a small description about how to use the maps can really help the users get started. But I couldn’t find any ways to add a text box in Carto.

Results and Reflection

Although I am quite happy about how these visualization turned out, I feel there is still a lot of work to be done. Running predictive analysis with different models can further help us perfect the results of the visualization and aid our efforts in eradicating the COVID-19 virus.

Carto is a really powerful tool to work with, but is pretty limited in terms of the flexibility it offers. For instance there was no way for me to add the text box explaining the map terminology I used. I also couldn’t adjust the position of the widgets on the final maps.

I also wanted to use different projections for the states of Alaska and Hawaii, but the process documented by Carto to make it possible was quite vague and couldn’t be implemented properly. Overall I believe the visualizations turned out well and really fulfilled their purpose of helping user understand the COVID data more accurately and efficiently.

Going further with this project, we can use different predictive analysis models and compare the result to see which model more accurately the number of cases. We can also compare the results produced by various models to the actual projected results, which would help us train the model for improved accuracy. We can also run the same use case on the World COVID Data, to identify and isolate potential hotspots. In this age of technology, readily available data can help us tackle a pandemic in a much better way.