Murals, Messages, and Moods

Mean, Median, and Moose

Today on Mean, Median, and Moose you’ll be treated to data we collected or generated all by ourselves!

Asking ChatGPT to Create Some Sample Data

I have been playing around with ChatGPT and a few other online AI tools. I wondered how well it would replicate re-world data and preference at a population level. I wondered how closely it could replicate election results by randomly generating a series of poll responses.

First thing, I did was I had it view election coverage from the Windsor Star, CBC Windsor, CTV Windsor for the 2021 Election. I copied the links to ChatGPT asked it to summarize each story and made sure that it noted how parties performed in Windsor West.

I asked Chat GTP to review a number of news stories that I found by googling Federal Government News Windsor 2024, against asking ChatGPT to summarize the stories and not the important policy issues highlighted in them.

The top political issues in Windsor, Ontario, include:

1. Healthcare: Improving access and funding for healthcare services, especially mental health care and home care services.
2. Housing: Increasing the supply of affordable housing, implementing rent controls, and addressing homelessness.
3. Economic Development: Boosting local industries, including the auto sector, and improving job opportunities.
4. Education: Enhancing funding for schools and addressing educational disparities.
5. Environmental Policies: Promoting renewable energy projects and sustainable development initiatives

Based on these economic priorities and past election results, I asked ChatGPT to create a fake dataset of 250 responses of a poll about the riding of Windsor West in Windsor Ontario Canada including respondents: age, gender, level of education and household income level. To give additional contest on the demographic figures I had ChatGPT summarize Windsor West Riding Profile for each of the respondent categories.

Use past elections, survey data and local news as a basis to create the responses to the following questions:

Question 1: Did you vote in the last federal election?

Question 2: Which party did you vote for?

• Liberal
• NDP
• Conversative
• Green Party
• People Party of Canada

Question 3: How likely are you to vote in the next election?

• Very Likely
• Likely
• Neutral
• Unlikely
• Very Unlikely

Question 4: What is your top issue in the next election?

Although it offered a Python output that stumped me and I was lazy, so I asked for a CSV file. ChatGPT out 250 rows of data like this.

Respondent ID,Age,Gender,Education Level,Household Income,Voted in Last Election,Party Voted For,Likely to Vote Next Election,Top Issue

1,45,Male,Bachelor’s Degree,$50,000 – $75,000,Yes,Liberal,Very Likely,Economy

2,34,Female,High School Diploma,$25,000 – $50,000,No,N/A,Likely,Healthcare

3,29,Male,Master’s Degree,$75,000 – $100,000,Yes,NDP,Neutral,Environment

4,54,Non-binary,Some College,<$25,000,Yes,Conservative,Very Likely,Education 5,62,Female,Associate Degree,>$100,000,No,N/A,Unlikely,Housing

You can view the data here.

Created Sample of 253	Conservative	Green Party	Liberal	N/A	NDP	People’s Party of Canada
Count of Party Voted	35	34	37	111	35	1

Based on ChatGPT review of history and estimate of the future we could expect 57% turnout in Windsor West in the next Election up from the 43% in 2021.

Created Sample of 142 Voters	Conservative	Green Party	Liberal	NDP	People’s Party of Canada
Vote Percentage	24.6%	23.9%	26.1%	24.6%	<0.1%

Now this is an unweighted sample and digging through cross times, find significantly over samples higher levels of education and incomes for the Windsor West riding. The top issues also made me laugh for a bit.

Row Labels	Conservative	Green Party	Liberal	N/A	NDP	People’s Party of Canada	Grand Total
Economy	1		9		1	1	12
Education	21	4	17	11	7		60
Environment	2	7	3	42	6		60
Healthcare	5	19	3	16	19		62
Housing	6	4	5	42	2		59
Grand Total	35	34	37	111	35	1	253

Conservatives don’t care about the economy, Greens don’t care about the Environment, only people who didn’t vote want housing. Seems an accurate representation of my riding (sarcasm).

Collecting Personal Data (Katie)

I’ve been an avid user of Daylio, a mood tracking app, for over 5 years now. It started at a time when I felt a lot of anxiety and unrest, so I wanted to be more mindful of my moods. I’ve tracked my range of emotions every day for years, with a reminder on my phone popping up every 3 hours to choose my mood and activities I’ve been doing. Gradually, I noticed my moods transition from more negative to positive, and I mostly select “Good” as my mood these days.

A little over a month ago, I was faced with a new reason to collect personal data like this – I had noticed my fatigue, a symptom of my multiple sclerosis, seemingly impacting me more and more. But how could I really be sure without any data to back this up? And if it was happening, were there any patterns I could recognize to help with it? So, I transitioned my Daylio mood tracking to fatigue tracking, changing the 5 mood levels to 5 fatigue levels instead: Exhausted, Fatigued, Neutral, Awake, and Energetic. I also added activities like “Meal”, “Coffee”, and “Snack” to see if there was a correlation between my fatigue and when I was eating or drinking coffee.

Daylio offers a wealth of statistics and charts once you’ve been tracking for a week or more. While I only have fatigue data for the month of May so far, I took a look at the stats Daylio has to offer to see if my assumption that I’m feeling fatigued often is true, and to see if there was any correlation with the activities I had listed, which you can see below.

In May, I made 244 entries in Daylio (approximately 8 entries per day, with an entry about every 2 hours from 7am to 9pm). My average fatigue rating was 3.1 (Neutral), with a total of 3 Energetic, 66 Awake, 139 Neutral, 34 Fatigued, and 2 Exhausted fatigue levels entered in the month.

This told me that while I wasn’t doing terribly with my fatigue, I also wasn’t doing as well as I wanted. Ideally, I’d be “Awake” or “Energetic” over 50% of the time, not 28% of the time. I entered “Neutral” so often that Daylio considered my fatigue quite stable though, scoring me an 86/100 in their stability chart.

So, now I knew I was probably more fatigued than I’d like to be, and I took a look at the activities I had logged to see how these might be impacting me. According to Daylio’s “Most Influential Activities” chart, I was most awake after having a snack, shopping, having coffee (no surprise there!), when I was at work (much more surprise there), and after having a meal. Going for a walk, watching a movie, going to bed, traveling, and seeing my family were activities aligned with poorer fatigue levels.

Now, here’s where I had to take some of this with a grain of salt. While I could see snacks, shopping, coffee, and meals helping me feel more awake, “Work” likely appeared on the list simply because it was my most logged activity at 98 entries, and I also almost always have a meal and coffee while at work. I could see this when clicking on “Work” as an activity in Daylio’s stat tracker and viewing “Related Activities” with “Meal” occurring 22 times on the same day as work and there being a 71% relation between the two, and similarly with “Coffee”, at 67%.

Similarly and intuitively, I know “Going to bed” made the negative list since of course I’m more tired right before bed, “Movies” appeared negatively since I almost only watch a movie right before bed. In the same way, “Family” made the list as I always go to see my family Monday night right after work, when I’m naturally more tired.

All in all, tracking my fatigue in this way helped me be more mindful of it, confirmed my average fatigue level, and helped me see that eating and a cup of coffee are ways I can boost my energy level, even if only temporarily, whereas exercising through a walk might not be as energy-boosting for me as it is for others. Tracking your own personal data can be simple with the wide variety of tracking apps out there now, and it’s a great way to get to know yourself better, increase your mindfulness, and tackle personal goals in a highly intentional and analytical way.

Developing a Walking Tour

A few years ago, Doug’s company Parallel 42 Systems built a government-funded walking tour app. P42 requested and was granted the permission to commit the code written for this project to the commons. It’s called Pytheas, and since the code is free for anyone to use, they’ve been using it! Last year P42 collaborated with local LGBTQ+ activists on a tour of sites of historic significance to the queer community, and this year their gift to our local community is a curated cross-border art tour focused on murals.

The code is mostly written, so implementing a new Pytheas tour is all about data collection.

There are some data points needed before a mural can be included in the tour;

• Geographic coordinates
• A photograph of the mural
• The address of the nearest building

Ideally, the data for a mural includes the following information;

• The name of the artist
• A brief biography of the artist
• A description of the work, its origin, and its meaning

Murals are inherently ephemeral, and generally not well-documented. Data collection involved automated and manual steps. To get started, the team looked for existing data sets of murals in the two cities.

In Windsor, the Free for All Walls Festival in 2023 added dozens of murals to local streets. This project has an excellent website and map which provide a good starting point, and crucially for this project, artist information for each of the murals.

In Detroit, the Visit Detroit Mural Guide and the City’s Mural Map also provided some useful hints, though the City of Detroit’s map does not surface many of the key pieces of information we needed.

Starting with this seed data, Doug and the P42 team next determined the desired route for each tour based on background knowledge of each city and the objective of the tour, which is to promote cross-border tourism in the region.

P42 used social media to ask residents of each city their favourite murals, documented them along with the murals from the seed data that appear on our target route.

The next step was the development of a complete list of candidate murals. This work was performed by a pair of site surveys. The first survey was conducted via Google Maps Street View. Street View was used to find murals, and to understand the immediate environment around the murals.

At this point, P42 had a list of about sixty candidate murals on either side of the border that fit the basic criteria. The spreadsheet of candidate mural locations was geocoded with Geoapify, which is a service we’ve talked about on this show before. That geocoded spreadsheet was uploaded to Google Maps for use by the photographers hired to walk the routes and get photos for the tour.

P42’s photographers used telemetry tools to capture a GPX document identifying the locations of the murals, and performed on-the-spot curation of the murals and the tour by being the first ones to walk it.

Using QGIS, P42 converted the GPX files to CSV, hand-modified them to reflect the structure and naming convention of the geoJSON files that power Pytheas, and finally converted them to geoJSON. Supplementary information from various sources was manually added to the final geoJSON document.

You can see the results at https://motownmurals.tours. If you’re not local to Windsor/Detroit, you’ll be too far away to take the tour, but you can browse the list of murals and check out the photos.

The MMM group chat

Behind the scenes at Mean, Median, and Moose there’s a rollicking instant message group we use a little bit to coordinate the show, but mostly to post funny tweets, memes, and complaints about local politics. Given that we’ve been doing this for a few years, we wondered what kind of data the group chat itself could provide. To limit the scope a bit we analyzed all our messages from 2023 – all 29,272 of them. That’s an average of 80 messages a day. 20 messages per person per day….

Broken down by chat participants we see the most messages come from John, then Doug, Frazier and Katie in that order.

John is also the most prolific link sharer, but this time Frazier comes in at number 2:

What about the time of the message? We can see that May and June were the most popular months for posting:

December is the least popular – I think likely because this is when the Mean, Median, and Moosers might be busy with the holidays. Ironically some of our most popular episodes and posts are the Christmas specials.

Finally, we can’t do an analysis without a heatmap, so here’s where the Mean, Median, Moosers were most active according to day and hour. The weekdays below start on Sunday (numbered as 1).