This week’s Makeover Monday was in collaboration with the UN SDG Action Campaign in occasion of the upcoming Global Goals week. Positive Impact Events collected data through a survey which asked people to propose specific actions that the global event industry can take to help the effort towards achieving the Sustainable Development Goals set by the UN in 2015. More specifically, the survey asked people to pick what they think are the 5 most crucial goals among the total 17, and then suggest possible actions and their duration relatively to each of the chosen goals. The survey also collected data about the voter’s country and city of origin, their age, gender, education level, disabilities, and job role. The original visualisation was made in Google Data Studio and it looks like this:
My mistake at the beginning of this Makeover Monday was to try and fit too many things in the same dashboard. Between the top actions per goal, the proposed duration for those actions, and the choice of goals per country and population group, there is a lot of information to draw from this survey. However, to fit it all into one visualisation means overloading your viewer and giving them a bad time.
Instead, I chose to focus on a specific angle: how did people from different countries, age groups, genders, education levels etc. prioritise the different goals? This information is conveyed by just two graphs, a world map indicating the top goal voted by each country, with the bubble size indicating the number of voters, and a scatter plot indicating what percentage of each voter group selected each goal as one of their top 5.
The viewer can hover over each data point to see a short description of the voter group and their preferences to better understand the smaller details of the results, but overall they should be able to quickly get a general idea of which goals people find most important, and how this changes with different voter groups.
I spent about three days on this one and I am happy to say I have learned a few new things.
# load the tables
actions = pd.read_csv("Positive Impact Events - Actions.csv")
goals = pd.read_csv("Positive Impact Events - Goals.csv", header=None, names = ["Goal", "Goal name"])
# left join the tables (in my head, "pie" stood for "positive impact events". don't judge me.)
pie = actions.merge(goals, how='left')
# fix the case
columns_to_change = ["Country", "Goal name"]
for col in columns_to_change:
pie[col] = pie[col].str.title()
INDEX() would work in a normal table, but once I switched to a map visualisation Tableau would show the top alphabetical goals instead of the most voted goals. In the end, I got so frustrated that I re-opened python, re-loaded the data and created a separate dataset for each country, its mode, and the number of votes it received (to retain the bubble size), and used it to create the map visualisation. This is the code I used:# number of votes per goal per country
top_goal_count = pie.groupby(["Country"])["Goal name"].value_counts().to_frame("Votes")
# keep only top value per country
top_goal_count = top_goal_count.groupby(["Country"])["Votes"].nlargest(1).to_frame()
Note: for some reason, this code results in a dataframe with two “Country” columns. I have no idea why nor how to not make it do that. If anyone can tell me what I am doing wrong I would wildly appreciate it. Either way, it works, even if in a dirty way.
Further note: some countries are multimodal, i.e. there is more than one most popular goal. For those situations I simply let Python pick any one of them, since this was mostly caused by a very low number of voters, who all chose different goals.
Unfortunately, using a separate dataset for the map meant that I could not use it as a filter for the scatter plot. It would have been cool to see the composition of voters of different countries, but because they were separate tables, I did not know how to make that happen.
Oh boy am I glad this is done. Because of the amazing opportunity to have our visualisations shown at the UN General Assembly, I ended up spending multiple days on this, trying to make it interesting and appealing. As I mentioned, at first things were way too cluttered, with me trying to show off all the cool insight I had found in a single dashboard. It kept looking horrible. Luckily I realised I needed to take a step back, cut some information away, and stay minimal. Another big thanks to @AdamMico1 for helping me style the horrible mess that this dashboard was when I first contacted him, and to my friends who ended up being my rubber ducks during these past few days. Time to rest now!