Gentrification in New York: A Redesign of Manipulative and Informative Charts

Figure 1: A before and after example of gentrification in new York [1]

Intro

New York is often a key city referenced in the discussion of gentrification in urban neighborhoods. The initial goal of this project was to use census data, and information visualization principals, to both inform and manipulate the viewer about the impacts of gentrification in New York's main Burroughs. From the initial design process came a set of 4 charts, 2 informative and 2 manipulative. The purpose of this redesign, however, was to revisit the charts made during the initial design process, and make improvements based on feedback, and new skills learned.

Figure 2 (left): An example of a manipulative chart from the first design process with respect to New York’s population change 2000–2017. Figure 3 (right): an example of one final redesign that was made for Figure 2

The Initial Design

Critique

For the initial design, the group chose to split the charts into 2 different stories where we would focus on separate impacts of gentrification. As such, 1 manipulative and 1 informative chart were used to represent both changes in race in New York, and then changes in income and home values from 2000 to 2017.

The first design the group created was a set of informative pie charts that showed the population differences in New York from 2000 and 2017:

Figure 4: First final design of story 1, informative chart showing the population changes in New York in 2000 and 2017

While pie charts are an easy to interpret method for representing proportions [2], the visualization suffered from a handful of improper design decisions:

1. The use of different sizes to represent the change in total population is not something that can be easily perceived, particularly with circles [3].
2. Pie charts generally fail at communicating change efficiently [2], so the differences in specific populations by race over time, were not well represented.
3. Using a tooltip to show the values instead of just directly labeling the slices, reduced accessibility to the chart
4. Although color encoding was used, and a tool tip said what race each slice was, no legend was included to show which color represented which race. This decision again reduced accessibility and also created some ambiguity.

For the second design (Figure 5), we again used a pie chart to tell our story, but redacted some of the information to add to the manipulative aspect of the design. However, the level of manipulation in the chart suffered as it was limited solely to underrepresenting the data. As such, we could have further mislead the viewer with additional details, like misleading text and titles, or adding more nuanced manipulation through further data aggregation, and visual representations [4]. The chart also suffered from some similar issues as our other pie chart (Figure 4) with respect to accessibility, and legends for the color encodings.

Figure 4: First final design of story 1 for a manipulative pie chart of population change

For the third design, and second informative chart, we showed changes in home value and income using a random sampling scatterplot for both gentrified and non gentrified neighborhoods:

Figure 5: First final design for story 2 informative chart showing home value and income changes from 2000 and 2017 in both gentrified and non gentrified areas

The third design was successful at enhancing clarity with respect to categorical color encoding between charts, such that non gentrified was orange, and gentrified was green, and it was labeled as such [5]. It also did well to represent a range of values in the data using a random sampling. However, the chart ended up being unintentionally manipulative as the two plots were on different scales, but the home value chart’s upper outlier skewed the data to the left. Because of this, at a glance, the home values appear to have not increased as dramatically as the income [6], which of course, upon closer inspection, is not the case.

For the fourth and final design, we created a manipulative bar chart for the median home values and income changes:

Figure 6: First final design for story 2 manipulative chart showing home value and income changes from 2000 and 2017 in both gentrified and non gentrified areas

A number of design decisions made this chart our more successful manipulative chart:

1. The bars were ordered from greatest to least for home value, and least to greatest for income, thus prompting the viewer to perceive trend lines which do not actually exist [2].
2. Omitting color encodings for gentrified and non gentrified data deliberately added ambiguity to the chart, potentially leading the viewer to make incorrect assumptions, such as, thinking the leftmost bar in one chart represents 2000, and the rightmost 2017 [6].
3. The title added more questions, rather than provide a clear answer, and also used divisive wording to influence the viewers perceptions.
4. The spread of the values was manipulated in the aggregations to appear smaller. For instance, in gentrified New York, New York the change in home values had increased by nearly 700%. But, by using the average of the proportional change, we were able to display the chart with apparently much smaller values. Additionally, the scaling of the y-axis was deliberately altered to make some bars appear taller, and others appear shorter. All of this then helped to nudge the viewer into following the narrative of the title [6].

While there were a number of successful design decisions with the final chart, it did fall short with accessibility. Specifically, the labeling of the x-axis displays the names of the Burroughs vertically, thus making the labels difficult to read.

The Design Process

Sketches

After critiquing the initial designs, a new set of preliminary sketches were drawn up. From the first design of the informative pie chart, the following redesigns were created:

Figure 7 (left), 7.1 (top left), 7.2 (top right), 7.3 (bottom): sketches showing NY changes in demographics form 2000 to 2017. Figure 8 (right): sketch of interactivity for figure 2.2

Because people are better at perceiving change through line and bar charts [3], and making comparisons on a common scale, some sketching exploration was done in Figure 7 to pull the data out of the initial pie chart format and into something more intuitive. Additionally, figure 7.3 looked at trying to reduce, or simplify the amount information the viewer needed to interpret. Figure 8 then considered adding interactive elements to allow the viewer to further explore the data.

For the redesign sketch of the manipulative pie chart, more misleading detail was planned:

Figure 8(left): Redesign of the manipulative pie chart to include a more misleading title, and the addition of more misrepresented data to sway the viewer. Figure 9(right): Alternative methods for displaying the changes in population from the manipulative pie chart

For Figure 8, a more divisive title was included, and a plan was laid out to use text callouts for the decline in White population to further manipulate the viewer. Figure 9 then looks at an alternative design to the pie chart that adds back in some data showing the decline in White and Black populations but the rise on others.

Figure 8 shows some alternate ideas to the initial scatterplot from Figure 5 which would allow for there to be more data included in the chart:

Figure 10: alternatives to the chart in figure 9 that include a broader picture of the data

Ultimately, the bottom scatter plot with jitter was selected to move forward as to not completely toss out the original design, which was generally successful sans the outliers and opposing scaling.

As for the final manipulative bar chart from Figure 6, because the amount of changes required seemed to be limited to the tilting of the x-axis labels, no sketch was created, as a complete redesign of the chart did not seem necessary.

Prototypes and Testing

For the first chart — redesign of the initial informative pie chart — the bar chart sketch was selected to show the changes in races from 2000 to 2017 (Figure 11). Having a side by side comparison on a similar scale, rather than using pie charts, made the comparison of overall change a bit easier to interpret. Additionally, having the color encoding legend available, helped to see what race corresponded to what section of the bar. However, feedback was received that, while the tool tip helped a little, quickly understanding the change of population of one race was difficult as they couldn’t tell if one bar was actually increasing, or if that was just the total population increasing. Additionally, the tester noted that Native and Pacific Islander populations were too hard, if not impossible to see, and access with a tool tip.

Figure 11: First prototype for the informative NY population chart, redesigning Figure 4 the informative pie chart

To adapt the bar chart with the feedback received, another prototype was created using the alternate sketch from Figure 7.1 combined with the interactivity design from Figure 8:

Figure 12: Iteration for informative NY population chart, redesigning Figure 4 the informative pie chart

The second prototype (Figure 12) for the first informative chart, incorporated the feedback to make the changes in population between years a bit more clear through the slopes of the lines. Interactivity was also added to allow the viewer to filter the data between total population change, and the changes in gentrified, and non gentrified areas. However, additional feedback was received that the dots were still difficult to select with the tool tip, and the Native and Pacific Islander dots were still tricky to access as they had some overlap. The feedback received was then incorporated to create the final design in the next section.

For the manipulative population chart, the critique from the first design was used to add more deception:

Figure 13: First prototype for the manipulative NY population chart, redesigning Figure 5 the manipulative pie chart

For the prototype (Figure 13), the title and subtitle were selected to suggest that gentrification is not really an issue in New York, and instead the real issue is the influx of “immigrants” to the city, causing the White population to “plummet”. Which, while White residents in New York did decrease in total population, actual gentrified neighborhoods had a notable increase, as per the above prototype shown in Figure 12. The text also calls out the population change of White “residents” and the combined change of Asian and Hispanic “immigrants”. Additionally, other data was removed, specifically the decline in the Black population as it could potentially influence the focus on the decreasing White population. Finally, red was used to suggest some level of alarm through the encoding of the percent change in white population.

From the prototype in Figure 13, mostly positive feedback was received from testers, positive as in “why would you make such an divisive chart?”. As such, only some minor changes were made for the final design of the chart.

For the second informative chart on home value and income change, the upper outliers were removed from the data, the two plots were put on the same scale, and jitter was added to allow for all the data points to be represented:

Figure 14: First prototype for the informative income and home value chart, redesigning Figure 6 the informative scatter plot

For the prototype in Figure 14, again mostly positive feedback was received, but some interactivity and design adjustments were added to the final design to allow for clear comparisons, again, discussed in the next section.

For the final design, the notes from the critique were incorporated to tilt the labels to enhance readability (Figure 15). Feedback was received that the charts were now a bit easier to read, and a suggestion was made to add some color encoding to just help distinguish between the two charts.

Figure 15: First prototype for the manipulative income and home value chart

Final Redesign

Combining the findings in the critique, along with the feedback received, updates and adjustments were made to the prototypes to create the final designs. For the first redesign of the informative chart seen in Figure 4, the following interactive visualization was made:

Figure 16: The final redesign of Figure 4: the informative pie chart

As mentioned in the critique, interpreting change, making comparisons, and accessibility were all notable concerns within the original pie chart design. As such, the following design decisions and improvements were made along with some additional updates:

1. Instead of requiring the viewer to try and make comparisons between slices of the pie chart, the lines allow the viewer to more quickly interpret rates of change [2]. Additionally, the tool tip tells the user the exact value of population change for a select race to help solidify their understanding of the data.
2. Accessibility and expressiveness was improved with the inclusion of a legend for the color encodings.
3. The dots on the chart were made larger so it was easier to hover over them, and interactivity was added so that the viewer could see the difference in lines/points which were overlapping. Both of which work to further improve accessibility.
4. Including a method for the viewer to filter data helped to create a more nuanced story about the impacts of gentrification, rather than just showing the total population changes.

For the second redesign of the manipulative pie chart shown in Figure 5, the following final bar chart was created:

Figure 17: The final redesign of Figure 5: the manipulative pie chart

As mentioned in the critique for the misleading pie chart, seen in Figure 5, the level of manipulation suffered as it was limited solely to underrepresenting the data. To redesign the chart, various good design techniques were combined with manipulative techniques to make the following design decisions:

1. Using the concept that “people’s perceptions of a visualization’s message, and their ability to recall it, are particularly influenced by the visualization’s title” [7], a divisive title and subtitle were created. Each worked to influence the viewers perception of the chart by suggesting “immigrants” were taking over New York, and gentrification is not an issue.
2. To compliment the title, the representation of the data was restricted by not separating the gentrified and non gentrified data. Actually making this separation would have shown an influx in White population to gentrified areas [7,8].
3. Data was also further restricted, and cherry picked to only show the decline in White population [8]. While the total Black population had also declined from 2000–2017, including that data in the chart had potential to detract from the focus of the story.
4. The text callouts for the population changes use the combined sum of new Asian and Hispanic residents to make the number appear larger and more “serious”. Additionally, the text callouts used highlighting “to ensure that the viewer will match the pattern in the [implied] data” [7].
5. Color encoding was used as a signal detection to imply alarm with respect to the decline in the White population in both the bars and the text callouts.
6. The idea that “viewers can immediately pull general statistics from the positions, lengths, … and intensities” [7] was used in a number of the above points as well, but it was also used to skew the scale of the y-axis to make the bar for the change in Hispanic population appear taller.
7. Arrow icons were also used to add movement and influence the idea of one population “plummeting” while another was “rocketing ”.

For the third redesign of the informative scatter plot from Figure 6, the following redesign was made:

Figure 18: The final redesign of Figure 6: the informative scatter plot

As mentioned in the critique, the scatter plot was unintentionally manipulative due to scaling and outliers. Additionally, there was a personal desire to include more data as to help tell a richer story. Because of this, the following design decisions were made:

1. To avoid skewing the home value data to the left, the upper outliers were removed.
2. Even with the outliers removed, the income change appeared to keep up with the home value change, as such, the scale for each plot was set to be equal to allow for more actuate comparisons [7].
3. Interactivity was also added to help better link the two charts, allowing the user to highlight a point or section of points to see where that cluster lies in the opposite chart. It is particularly helpful for seeing how the upper and lower bounds compare, and works to give the viewer a richer understanding of the data.
4. From the prototype, the sizes of the dots were also reduce and opacity was adjusted to help the viewer better see the distribution.

For the fourth redesign of the manipulative bar chart from Figure 7, the following redesign was made:

Figure 19: The final redesign of Figure 7: the manipulative bar chart

From the critique, and feedback the only issues to address for this chart was accessibility for reading the x-axis labels, and adding color encoding to help differentiate the data used in each chart. As such, the changes made for the chart were just tilting the labels, and changing the color for the home value chart to help differentiate the two charts encodings.

Takeaways

The three key takeaways:

1. The devil is in the details. It can be very easy to inadvertently mislead a viewer by underrepresenting data, not looking for accidental dark patterns, or cherry picking data to fit the pre determined narrative. For this particular project, unintended manipulation was an issue that arose with the preliminary design for the scatter plot. Having skewed data, and different scales, completely changed what the viewer understood about the chart at first glance. Because many people do not take significant time to inspect all the details of a chart, it can be easy to unintentionally change the intended story. This also leads to the next important takeaway: the need for feedback.
2. Feedback helps to provide an unbiased interpretation of a visualization. Often times we get caught up with the story we are trying to tell, or put more weight on an assumption that the viewer will fill in the missing pieces or nuances of a story. For the preliminary scatter plot chart, getting more initial feedback could have helped us realize how the information and story was being skewed. For the pie charts on the other hand, some assumptions were made that the viewer would mentally fill in some information, or that they would be guaranteed to interact with the tool tip. Getting more feedback from individuals outside of the class could have helped select better charting methods, and influence the addition of details to make the pie charts less ambiguous. It also helps to receive feedback from someone who is not familiar with the story or the data, because they can tell you their immediate interpretation of a chart without any influence or bias.
3. A desire to try new things, and be creative can negatively overshadow acceptable simplicity. While it is good to add creativity to visualizations, and try something new, it is still okay to use a simple, or obvious charting method. This is something that is notable in the informative pie charts as there was an attempt, and desire, to do something unique. However, it missed the mark on a number of basic design principals because it was simply not a good method for representing the data. As such, it is okay to use an obvious charting method like line charts to show changes over time, and then add in creativity through other avenues such as accessibility, color and size encodings, interactivity, highlighting and more.

References

[1] Image: https://petapixel.com/2014/04/04/pictures-ny-storefronts-document-decade-gentrification/

[2] 39 studies about human perception in 30 minutes. Kennedy Elliot, 2016.

[3] Lecture 6 slides, Perception. Evan Peck, 2024.

[4] Black Hat Visualization. Michael Correll, Jeffrey Heer.

[5] When to use quantitative and when to use qualitative color scales. Lisa Charlotte Muth, 2021

[6] The Good, the Bad, and the Biased: Five Ways Visualizations Can Mislead (and How to Fix Them), Danielle Albers Szafir, 2018.

[7] “The Science of Visual Data Communication: What Works”, Psychological Science in the Public Interest 2021, Vol. 22. Steven L. Franconeri, Lace M. Padilla, Priti Shah, Jeffrey M. Zacks, and Jessica Hullma, 2021.

[8] How People Actually Lie With Charts. Maxim Lisnic, 2023.