Recently the United Kingdom baked in a significant heatwave. With climate change being a real thing, an extreme heat event in the summer is not terribly surprising. Also not surprisingly, the BBC posted an article about the impact of climate change.
The article itself was not about the heatwave, but rather the increasing rate of sea level rise in response to climate change. But about halfway down the article the author included this graphic.
As graphics go, it is not particularly fancy—a dot plot with ten points labelled. But what this piece does well is using a dot plot instead of the more common bar chart. I most typically see two types of charts when plotting “hottest days” or something similar. The first is usually a simple timeline with a dot or tick indicating when the event occurred. Second, I will sometimes see a bar chart with the hottest days presented all as bars, usually not in the proper time sequence, i.e. clustered bar next to bar next to bar.
My issue with the the latter is always where is the designer placing the bottom of the bar? When we look at the best temperature graphics, we usually refer to box plots wherein the bar is aligned to the day and then top of the bar is the daily high and the bottom of the bar the daily low. It does not make sense to plot temperatures starting at, say 0º.
In this particular case, however, the dates would appear to overlap too closely to allow a proper box plot. Though I suspect—and would be curious to see—if the daily minimum temperatures on each of those ten hottest days have also increased in temperature.
As to the timeline option, this does a better job of showing not just the increasing frequency of the hottest days, but also the rising maximum value. In the early 20th century the hottest day was 36.7ºC, and you can see a definite trend towards the hottest days nearing and finally surpassing 40ºC.
I do wonder if a benchmark line could have been added to the chart, e.g. the summertime average daily high or something similar. Or perhaps a line showing each day’s temperature faintly in the background.
Finally, I want to point out the labelling. Here the designers do a nice job of adding a white stroke or outline to the outside of the text labels. This allows the text to sit atop the y-axis lines and not have the lines interfere with the text’s legibility. That’s always a nice feature to see.
Credit for the piece goes to the BBC graphics department.
The other day I was reading an article about the coming property tax rises in Philadelphia. After three years—has anything happened in those three years?—the city has reassessed properties and rates are scheduled to go up. In some neighbourhoods by significant amounts. I went down the related story link rabbit hole and wound up on a Philadelphia Inquirerarticle I had missed from early May that included a map of just where those increases were largest. The map itself was nothing crazy.
We have a choropleth with city zip codes coloured by the percentage increase. I was thrown for a bit of a loop as I immediately perceived the red representing lower values and green higher values, the standard green to red palette. But given that higher values are “bad”, I can live red representing bad and sitting at the top of the spectrum.
I filed it away to review later, but when I returned I visited on my mobile phone. And what I saw broadly looked the same, but there was a disconcerting difference. Take a look at the legend.
You can see that instead of running vertically like it did on the desktop, now the legend runs horizontally across the bottom. In and of itself, that’s not the issue. Though I do wonder if this particular legend could have still worked in roughly the same spot/alignment given the geographic shape of Philadelphia along the Delaware River.
Rather look at the order. We go from the higher, positive values on the left to the negative, lower values on the right. When you read the legend, this creates some odd jumps. For example, we move from “+32% to +49%” then to “+15% to +31%”. We would normally say something to the point of the increase bins moving from “+15% to +31%” then to “+32% to +49%”. In other words, the legend itself is a continuum.
The fix for this would be to simply flip the running order of the legend. Put the lower values on the left and then step up to the right. For a quick comparison, I visited the New York Times website and pulled up the first graphic I could find that looked like a choropleth. Here we have a map of the dangerous temperatures across the United States.
Note how here the New York Times also runs their legend horizontally below the graphic. But instead of running high-to-low like in the Inquirer, the Times runs low-to-high, making for a more natural and intuitive legend.
This kind of simple ordering change would make the Inquirer’s map that much better.
Credit for the Inquirer piece goes to Kasturi Pananjady and John Duchneskie.
Credit for the Times piece goes to Matthew Bloch, Lazaro Gamio, Zach Levitt, Eleanor Lutz, and John-Michael Murphy.
Editor’s note: I was having some technical issues last week. This was supposed to post last week.
Editor’s note two: This was supposed to go up on Monday. Still didn’t. Third time’s the charm?
Yesterday I wrote about a piece from the New York Times that arrived on my doorstep Saturday morning. Well a few mornings earlier I opened the door and found this front page: a map of the western United States highlighting the state of New Mexico.
Unlike the graphic we looked at yesterday, this graphic stretched down the page and below the fold, not by much, but still notably. The maps are good and the green–red spectrum passes the colour blind test. How the designer chose to highlight New Mexico is subtle, but well done. As the temperature and precipitation push towards the extreme, the colours intensify and call attention to those areas.
Also unlike the graphic we looked at yesterday, this piece contained some additional graphics on the inside pages.
These are also nicely done. Starting with the line chart at the bottom of the page, we can contrast this to some of the charts we looked at yesterday.
Here the designer used axis lines and scales to clearly indicate the scale of New Mexico’s wildfire problem. Not only can you see that the number of fires detected has spiked far above than the number in the previous years back to 2003. And not only is the number greater, the speed at which they’ve occurred is noticeably faster than most years. The designer also chose to highlight the year in question and then add secondary importance to two other bad years, 2011 and 2012.
The other graphics are also maps like on the front page. The first was a locator map that pointed out where the fires in question occurred. Including one isn’t much of a surprise, but what this does really nicely is show the scale of these fires. They are not an insignificant amount of area in the state.
Finally we have the main graphic of the piece, which is a map of the spread of the Calf Canyon and Hermits Peak fire, which was two separate fires until they merged into one. The article does a good job explaining how part of the fire was actually intentionally set as part of a controlled burn. It just became a bit uncontrolled shortly thereafter.
This reminded me of a piece I wrote about last autumn when the volcano erupted on La Palma. In that I looked at an article from the BBC covering the spread of the lava as it headed towards the coast. In that case darker colours indicated the earlier time periods. Here the Times reversed that and used the darker reds to indicate more recent fire activity.
Overall the article does a really nice job showing just what kind of problems New Mexico faces not just now from today’s environmental conditions, but also in the future from the effects of climate change.
Credit for the piece goes to Guilbert Gates, Nadja Popovich, and Tim Wallace.
Friday the Bureau of Labour Statistics published the data on the jobs facet of the American economy. Saturday morning I woke up and found the latest New York Times visualisation of said jobs report waiting for me at my door. The graphic sat\s above the fold and visually led the morning paper.
We have a fairly simple piece here, in a good way. Two sections comprise the graphic. The first uses a stacked bar chart to detail the months wherein the US economy lost jobs during the previous two and a half years. We can take a closer look in this second photo that I took.
Here we can see the stacked bars pile up with the most recent bars to the right. Some of the larger bars have labels stating the number of jobs either lost (top) or gained (bottom). I’m not normally a fan of stacked bar charts, because they don’t allow a reader to easily discern like-for-like changes. In this instance, the goal is to show how close all the little bits have come towards making up the three negative bars. Where I take issue is that I would prefer the designers used some sort of scale to indicate even a rough sense of how many jobs the various bars represent.
That issue crops up again to a slightly lesser degree with the bottom set of graphics. These compare the growth of hourly earnings and inflation both from February 2020. During the first few months of the pandemic and its recession, you can see earnings for those most directly impacted by shutdowns drop. But there is no negative scale accompanying the positive scale and that makes it difficult to determine just how far earnings fell for those in, say, leisure and hospitality.
The second part of the graphic works overall, however it’s just some of the finer design details that are missing and take away from the graphic’s overall effectiveness.
This all fits part of a larger trend in data visualisation that I’ve been noticing the last few months. Fewer charts seem to be using axes and scales. It’s not a good thing for the field. Maybe some other day I’ll write some things about it.
For this piece, though, we have an overall solid effort. Some different design decisions could have made the piece clearer and more effective, but it still does the job.
Credit for the piece goes to Ben Casselman, Ella Koeze, and Bill Marsh.
I noticed an interesting thing this morning. Over the holiday weekend I bookmarked a BBC News article about new airlines because it included a small graphic showing the number of airlines started during the pandemic (32) and the number of new airlines lost during the pandemic (55). The graphic used a stock three-dimensional illustration of a passenger airlines with a blank white body. From the top of the body rose two white bars, next to the left was the shorter of the two with a 32. The right was taller and had a 55. Above each was a header saying something to the effect of “Airlines started in 2020” and “Airlines lost in 2020”, respectively. Funny thing this morning that when I returned to the bookmark with this post in mind, the article’s graphic had disappeared.
This weekend I happened to start re-reading 1984, George Orwell’s classic dystopian novel about a man named Winston Smith. He works in the Records Department and is tasked with “rectifying” misstatements. I had just finished reading the section where Orwell describes Smith’s work wherein he takes previously published newspaper articles about statistics and figures and then edits them to include new numbers aligned with the actual outputs. This way should anyone read the old article for evidence of a previous past, they find the output forecasts have always been correct. He then destroys the written record of the old past by dumping it into a memory hole, a pneumatic tube that delivers it straight to a furnace where the old past is incinerated and thus replaced with Smith’s new version.
When I read the article again, because the graphic was gone, I read a paragraph that had figures for 2021. I cannot recall those numbers being present earlier this weekend. But they are roughly where I remember the old graphic being. Yet the article includes no note about any edits to a previous version let alone what those edits may have been. And so now I am left wondering if I really saw what I think I remember that I saw. How very Orwellian.
But let’s assume I did see what I thought I saw, the graphic was actually unnecessary. It presented two figures, 32 and 55. The bar chart itself had no axis labels and that made it a bit difficult to believe the numbers themselves. It did not help that the white bars blended almost seamlessly into the white body of the airliner. Moreover, the graphic was large and fit the full width of the text column. For two figures.
My initial goal was to show this graphic I made to show just how little space truly needs to be used to show an effective graphic. I also changed the direction of the bars. Instead of making one bar about the positive change and the other the negative change, I made both bars about the change. Therefore the one bar moved upwards with the positive (32) and the other downwards with the negative (55). I then plotted a dot to show the net change between the two. Yes, 32 airlines were created in 2020. But that still made for a net loss of 23 that year.
But because the graphic was missing and there was some new text for 2021 figures, I decided to incorporate them as well to show how the trend basically continued year over year.
I left the white space to the right to illustrate how you really do not need a full-width graphic to display only six data points, itself a three-fold increase on the original graphic’s data content. The original graphic contained more illustrated plane than it did data content.
Graphics should be about the data, not about the splashy, flashy, whizbang background content that ultimately distracts our attention away from what should be the focal point of the piece: the data. The article still contains photos of planes with the livery of the new airlines, of empty terminals to represent the pandemic losses, and portraits of executives. This graphic did not need an illustrated plane taking over the graphic. It needed to only show those two numbers.
I would even contend that the article could have made do with a simple factette, two big numbers. Airlines closed in 2020 and the airlines opened. It need not be fancy, but it quickly delivers the big numbers with which the reader should be concerned. You don’t need to see an aircraft or a terminal. You could add some colour to the numbers or even a minus sign as there is a significant difference between a 55 and a -55. But all in all, the graphic need not be full width like it was originally.
But I think we should all keep in mind the value of transparency. The graphic did exist, of that I am certain. But future readers or even my sanity cannot be sure that it did. And in an era where “fake news” and fact-checking are important, I wonder if we need to be including corrections notes in more of our news articles. Because if we lose faith in our news, we have little left to lean upon in our societal discourse about the events of our time.
Yesterday I focused on the big graphic from the New York Times that crossed the full spread of the front/back page. But the graphic was merely the lead graphic for a larger piece. I linked to the online version of the article, but for this post I’m going to stick with the print edition. The article consists of a full-page open then an entire interior spread, all in limited colour. The remainder of the extensive coverage consists of photo essays and interviews that understandably attempt to humanise the data points, after all, each dot from yesterday represented one individual, solitary, human being. That is an important element of a story like this and other national and international tragedies, but we also need to focus on the data and not let the emotion of the story overwhelm our rational and logical analysis.
From a data visualisation standpoint the first page begins simply enough with a long timeline of the Covid-19 pandemic charting the number of absolute deaths each day. As we looked at yesterday, the absolute deaths tell part of the story. But if we were to have looked at the number of absolute cases in conjunction with the deaths, we could also see how the virus has thus far evolved to be more transmissible but less lethal. Here the number of daily deaths from Omicron surpassed Delta, but fell short of the winter peak in early 2021. But the number of cases exploded with Omicron, making its mortality rate lower. In other words, far more people were getting sick, but as far fewer were dying.
An interesting note is that if you take a look at the online version, there the designers chose a more stylised approach to presenting the data.
Here they kept the dot approach and simply stacked and reordered the dots. However, I presume for aesthetic reasons, they kept the stacking loose dots and dropped all the axis lines because it does make for a nice transition from the map to this chart. But they also dropped all headings and descriptors that tell the reader just what they are looking at. These decisions make the chart far less useful as a tool to tell the data-driven element of the story.
There are three annotations that label the number of deaths in New York, the Northeast, and the rest of the United States. But what does the chart say? When are the endpoints for those annotations? And then you can compare the scale of the y-axis of this chart and compare it to the printed version above. A more dramatic scale leads to a more dramatic narrative.
This sort of visual style of flash and fancy transitions over the clear communication of the data is why I find the print piece more compelling and more trustworthy. I find the online version, still useful, but far more lacking and wanting in terms of information design.
The interior spread is where this article shines.
From an editorial design standpoint, the symmetry works very well here. It’s a clear presentation and the white space around the graphic blocks lets that content shine as it should in this type of story. Collectively these pieces do a great job telling the story of the pandemic thus far across the nation. The graphics do not need a lot of colour and make do with sparse flash. Annotations call the reader’s attention to salient points and outliers.
From a content standpoint, I would be particularly curious if we have robust data for deaths by education level. Earlier this year I recall reading news about a study that said education best correlated to Covid cases, and I would be curious to see if that held true for deaths. Of course these charts do a great job of showing just how effective the vaccines were and remain. They are the best preventative measure we have available to us.
Here I disagree with the design decision of how to break down the states into regions. The Census Bureau breaks down the United States into four regions using the same names as in the graphic above. However, if you look closely at the inset map, you will see that Delaware, Maryland, and West Virginia in particular are included as part of the Northeast. (I cannot tell if the District of Columbia is included as part of the Northeast or South.)
Now compare that to the Census Bureau’s definition:
If you ask me to include Delaware and Maryland as part of the Northeast, well, if you’re selling it, I’ll buy it. After all, just because the Census Bureau defines the United States this way does not mean the New York Times has to. Both are connected to the Northeast Corridor via Amtrak and I-95 and are plugged into the Megalopolis economy. Maybe the Potomac should be the demarcation between Northeast and South. But I struggle to understand West Virginia. Before you go and connect it to the Northeast, I would argue that West Virginia has far more in common with the Midwest geographically, economically, and culturally.
More critically, given this issue, it strikes me as a serious problem when the online version of the chart—with the aforementioned issues—does not even include the little inset to highlight this at best unusual regional definition.
And so while I have reservations about the data—how would the data have looked if the states were realigned?—the design of the line charts overall is good.
Again, I am talking about the print version, not that online graphic. I would argue that the above screenshot is barely even a chart and more “data art” or an illustration of data. Consider here, for example, that for the South we have that muted slate blue for the dots, but the spacing and density of the dots leads to areas of lighter slate and darker slate. But a lighter slate means more space between stacked dots and darker slate means a more compact design. A lighter colour therefore pushes the “edge” of the line further up the y-axis and artificially inflates its value, not that we can understand what that value is as the “chart” lacks any sort of y-axis.
Finally the print piece has a set of small multiples breaking down deaths by income in the three largest American cities: New York, Los Angeles, and Chicago. These are just great little charts showing the correlation between income and death from Covid, organised by Zip code.
But this also serves as a stark reminder of just how much better the print piece is over the online version. Because if we take a look at a screenshot from the online article, we have a graphic that addresses all the issues I pointed out earlier.
I am left to wonder why the reader of the online version does not have access to this clearer and more accurate representation of the data throughout the piece?
To me this article is a great example of when the print piece far exceeds that of the online version. Content-wise this is a great story that needed to be told this weekend, but design wise we see a significant gap in quality from print to online. Suffice it to say that on Sunday I was very glad I received the print version.
Credit for the piece goes to Sarah Almukhtar, Amy Harmon, Danielle Ivory, Lauren Leatherby, Albert Sun, and Jeremy White.
Earlier this week I read an article in the Philadelphia Inquirer about the political prospects of some of the candidates for the open US Senate seat for Pennsylvania, for which I and many others will be voting come November. But before I get to vote on a candidate, members of the political parties first get to choose whom they want on the ballot. (In Pennsylvania, independent voters like myself are ineligible to vote in party primaries.)
This year the Republican Party has several candidates running and one of them you may have heard of: Dr. Oz. Yeah, the one from television. And while he is indeed the front runner, he is not in front by much as the article explains. Indeed, the race largely had been a two-person contest between Oz and David McCormick until recently when Kathy Barnette pulled just about even with the two.
In fact, according to a recent poll the three candidates are all statistically tied in that they all fall within the margin of error for victory. And that brings us to the graphic from the article.
Conceptually this is a pretty simple bar chart with the bar representing the share of the support of those polled. But I wanted to point out how the designer chose to represent the margin of error via hatched shading to both sides of the ends of the red bar.
In some cases the hatch job does not work for me, particularly with those smaller candidates where the bar goes negative. I would have grave reservations about the vote should any candidate win a negative share of the vote. 0% perhaps, but negative? No. I also don’t think the grey hatching works as well over the grey bar in particular and to a lesser degree the red.
I have often thought that these sorts of charts should use some kind of box plot approach. So this morning I took the chart above and reworked it.
Overall, however, I really like this designer’s approach. We should not fear subtlety and nuance, and margins of error are just that. After all, we need not go back too far in time to remember a certain candidate who thought she had a presidential election locked up when really her opponent was within the margin of error.
Everyone knows inflation is a thing. If not, when was the last time you went shopping? Last week the Boston Globelooked specifically at children’s shoes. I don’t have kids, but I can imagine how a rapidly growing miniature human requires numerous pairs of shoes and frequently. The article explores some of the factors going into the high price of shoes and uses, not very surprisingly, some line charts to show prices for components and the final product over time. But the piece also contains a few bar charts and that’s what I’d like to briefly discuss today, starting with the screenshot below.
What we see here are a list of countries and the share of production for select inputs—leather, rubber, and textiles—in 2020. At the top we have a button that allows the user to toggle between the two and a little movement of the bars provides the transition. The length of the bar encodes the country in question’s market share for the selected material.
We also have all this colour, but what is it doing? What data point does the colour encode? Initially I thought perhaps geographic regions, but then you have the US and Mexico, or Italy and Russia, or Argentina and Brazil, all pairs of countries in the same geographic regions and yet all coloured differently. Colour encodes nothing and thus becomes a visual distraction that adds confusion.
Then we have the white spaces between the bars. The gap between bars is there because the country labels attach to the top of the bars. But, especially for the top of the chart, the labels are small and the gap is at just the right height such that the white spaces become white bars competing with the coloured bars for visual attention.
The spaces and the colours muddy the picture of what the data is trying to show. How do we know this? Because later in the article we get this chart.
This works much better. The focus is on the bars, the labelling is clear, almost nothing else competes visually with the data. I have a few quibbles with this design as well, but it’s certainly an improvement over the earlier screenshot we discussed. (I should note that this graphic, as it does here, also comes after the earlier graphic.)
My biggest issue is that when I first look at the piece, I want to see it sorted, say greatest to least. In other words, Furniture and bedding sits at the top with its 15.8% increase, year-on-year, and then Alcoholic beverages last at 3.7%. The issue here, however, is that we are not necessarily looking at goods at the same hierarchical level.
The top of the list is pretty easy to consider: food, new vehicles, alcoholic beverages, shelter, furniture and bedding, and appliances. We can look at all those together. But then we have All apparel. And then immediately after that we have Men’s, Women’s, Boys’ , Girls’, and Infants’ and toddlers’ apparel. In other words, we are now looking at a subset of All apparel. All apparel is at the same level of Food or Shelter, but Men’s apparel is not.
At that point we would need to differentiate between the two, whilst also grouping them together, because the range of values for those different sub-apparel groups comprise the aggregate value for All apparel. And showing them all next to Food is not an apples-to-apples comparison.
If I were to sort these, I would sort by from greatest to least by the parent group and then immediately beneath the parent I would display the children. To differentiate between parent-level and children-level, I would probably make the bars shorter in the vertical and then address the different levels typographically with the labels, maybe with smaller type or by putting the children in italic.
Finally, again, whilst this is a massive improvement over the earlier graphic, I’d make one more addition, an addition that would also help the first graphic. As we are talking about inflation year-on-year, we can see how much greater costs are from Furniture and bedding to Alcoholic beverages and that very much is part of the story. But what is the inflation rate overall?
According to the Bureau of Labour Statistics, inflation over that period was 8.5%. In other words, a number of the categories above actually saw price increases less than the average inflation rate—that’s good—even though they were probably higher than increases had been prior to the pandemic—that’s bad. But, more importantly for this story, with the addition of a benchmark line running vertically at 8.5%, we could see how almost all apparel and footwear child-level line items were below the inflation rate. But the children and infant level items far exceeded that benchmark line, hence the point of the article. I made a quick edit to the screenshot to show how that could work in theory.
Overall, an interesting article worth reading, but it contained one graphic in need of some additional work and then a second that, with a few improvements, would have been a better fit for the article’s story.
Here’s an interesting post from FiveThirtyEight. The article explores where different states have spent their pandemic relief funding from the federal government. The nearly $2 trillion dollar relief included a $350 billion block grant given to the states, to do with as they saw fit. After all, every state has different needs and priorities. Huzzah for federalism. But where has that money been going?
Enter the bubbles.
This decision to use a bubble chart fascinates me. We know that people are not great at differentiating between area. That’s why bars, dots, and lines remain the most effective form of visually communicating differences in quantities. And as with the piece we looked at on Monday, we don’t have a legend that informs us how big the circles are relative to the dollar values they represent.
And I mention that part because what I often find is that with these types of charts, designers simply say the width of the circle represents, in this case, the dollar value. But the problem is we don’t see just the diameter of the circle, we actually see the area. And if you recall your basic maths, the area of a circle = πr2. In other words, the designer is showing you far more than the value you want to see and it distorts the relationship. I am not saying that is what is happening here, but that’s because we do not have a legend to confirm that for us.
This sort of piece would also be helped by limited duty interactivity. Because, as a Pennsylvanian, I am curious to see where the Commonwealth is choosing to spend its share of the relief funds. But there is no way at present to dive into the data. Of course, if Pennsylvania is not part of the overall story—and it’s not—than an inline graphic need not show the Keystone State. In these kinds of stories, however, I often enjoy an interactive piece at the end wherein I can explore the breadth and depth of the data.
So if we accept that a larger interactive piece is off the table, could the graphic have been redesigned to show more of the state level data with more labelling? A tree map would be an improvement over the bubbles because scaling to length and height is easier than a circle, but still presents the area problem. What a tree map allows is inherent grouping, so one could group by either spending category or by state.
I would bet that a smart series of bar charts could work really well here. It would require some clever grouping and probably colouring, but a well structured set of bars could capture both the states and categories and could be grouped by either.
Overall a fascinating idea, but I’m left just wanting a little more from the execution.
We’re starting this week with an article from the Philadelphia Inquirer. It looks at the increasing number of guns confiscated by the Transportation Security Administration (TSA) at Philadelphia International Airport. Now while this is a problem we could discuss, one of the graphics therein has a problem that we’ll discuss here.
We have a pretty standard bar chart here, with the number of guns “detected” at all US airports from 2008 through 2021. The previous year is highlighted with a darker shade of blue. But what’s missing?
We have two light grey lines running across the graphic. But what do they represent? We do have the individual data points labelled above each bar, and that gives us a clue that the grey lines are axis lines, specifically representing 2,000 and 4,000 guns, because they run between the bars straddling those two lines.
However, we also have the data labels themselves. I wonder, however, are they even necessary? If we look at the amount of space taken up by the labels, we can imagine that three labels, 2k, 4k, and 6k, would use significantly less visual real estate than the individual labels. The data contained in the labels could be relegated to a mouseover state, revealed only when the user interacts directly with the graphic. Here it serves as a “sparkle”, distracting from the visual relationships of the bars.
If the actual data values to the single digit are important, a table would be a better format for displaying the information. A chart should show the visual relationship. Now, perhaps the Inquirer decided to display data labels and no axis for all charts. I may disagree with that, but it’s a house data visualisation stylistic choice.
But then we have the above screenshot. In this bar chart, we have something similar. Bars represent the number of guns detected specifically at Philadelphia International Airport, although the time framer is narrower being only 2017–2021. We do have grey lines in the background, but now on the left of the chart, we have numbers. Here we do have axis labels displaying 10, 20, and 30. Interestingly, the maximum value in the data set is 39 guns detected last year, but the chart does not include an axis line at 40 guns, which would make sense given the increments used.
At the end of the day, this is just a frustrating series of graphics. Whilst I do not understand the use of the data labels, the inconsistency with the data labels within one article is maddening.