Though I’m not officially a graphic artist, some of my work might be categorized as belonging to that fine tradition. Here are some examples of toys, tools, or explorations I’ve done that have that presence.

Arranging the days in a year specifically for travel planning

     

I couldn’t find a simple enough calendar to draw on to manage possibly conflicting travel and conference dates, so I typeset one for myself.

It’s intentionally very plain looking to allow the scribbles, circlings, and annotations that will give it its functionality and visual life. Its layout was designed to keep few breaks (and only culturally meaningful ones) in the numbers—allowing circled date ranges to stay contiguous more often.

Note the distinction between pencilled-in conferences and ballpoint blue confirmed appearances in the marked example (the right hand image above); I am sure you’ll come up with a visual language that’s appropriate for your own needs, if you choose to print one out. Who needs sunsets and pinups when you can turn the meaningful variations in your own schedule into a satisfying visual pattern?

Artsy goofing around with typeface outlines

     

One Tuesday not long ago I downloaded the FreeType C library which allowed me to directly access outlines of fonts a generous soul at Adobe provided a few months before. I scaled each face to match a prototype (drawn in red) by comparing heights of upper case letters, then slid each character along the baseline until as little as possible was outside the overlap of the prototype and new characters. That Sunday I had some interesting ghosts to look at: comparisons of letterforms that hundreds of people spend untold amounts of time on; themselves transcribing written and printed forms that took millennia to evolve.

(Be careful clicking on that second thumbnail above: it’s linked to a 4 Mb PDF that draws 166 faces on top of one another. It brings Acrobat almost to its knees; but it’s great fun to zoom in really close and look at how the individual letters compare.)

I'm enthralled by the approximating traces—especially in the lower case ‘g’ in the oldstyle faces—it reminds me of Giacometti’s searching line. He was obsessed by finding the perceptual boundaries around physical objects, as I am around “thought objects” like letters. I’m amused by the upper case ‘J’ wagging its tail, and the ‘Q’ with a blur of support like Balla’s “Dynamism of a Dog on a Leash.” But I’m wondering how (whether?) this can support careful thought about what makes a character recognizable or gives type its almost human personality on the page.

In fact, people seem to recognize characters the same way they recognize faces: as a set of features, rather than the averaged-together gestalts that this exploration suggests. If you want to follow up on the real science of letter recognition, take a look at Denis Pelli’s fascinating work on the subject, especially Identifying Letters at the top of his publications page at the time of this writing.

Sparklines work with Edward Tufte

  

I worked with Edward Tufte a little to help illustrate his essay on “sparklines:” small, intense, word-sized graphics that live easily in a text-covered page or grid. This is a link to my page of sparklines work, if you’re interested in the raw output of my tiny graph generator (hacked from my “Visual Spreadsheet” work dating back to 1994).

LifeLines: a proposal for a low-affect monument

Detail: simulated image of Ubatuba granite with carved lines and names (refuge view)

Schematic drawing for front: facing the park from a small rise (prospect view)

Schematic drawing for back: forming a 15 foot wide passage with the park’s wall (refuge view)

I was twelve blocks away and watching when the World Trade Center towers collapsed and I came to terms with the enormity of the event in my own way, saddened and disgusted with the politicized media circus that followed within hours. Andy Silverman, a broker (at that time) on the floor of the New York Stock Exchange was even closer. When he suggested we enter the competition for a monument I tried to decline but he cornered me by saying “If you think you can make a more respectful one, it’s your responsibility to.”

A simple design came out of conversations with Andy and later Marc Pittsley. It was among the several thousand not chosen, but schematic sketches (made from the most of the actual names and birthdates) of the front and back were hung at InfoVis 03. I believe that data can sometimes speak for itself; that the role of the monument designer is to step out of the way of the pilgrim’s reflections; that most current must-be-explained design refers to more the architect than the subject. I believe that the stories people take home should be about the emotional closure they find; not shiny spikey glass and glowing lights. I'm afraid most don't agree, but at least Ben Shedd, a new friend from that InfoVis told me it literally brought tears to his eyes.

Turning a chart into a glyph to help people recognize similar market reactions

  

On certain days in the financial world, “information days,” important information is released—e.g., numbers that say whether a company hit its profit projections, or news that a company has moved to a higher D&B credit rating. If you knew how the market reacted to a certian kind of news, you would be better prepared to react yourself, making money by buying just before the market goes up, for example.

So the important issues to bring forward are: 1) when the information was released; 2) the shape of the market’s reaction (flat? up? down, then up ten minutes later?); and 3) whether that reaction was consistent enough to bet on. These steps in the analysis task turned into visual features in this way:

1) The exact moment of the visualization is shown as a vertical line, like all other moments in the tiny charts, but that line was drawn darker to be a visual landmark. Note (click to see a larger image) that the time slices increase in size as the day goes on, since the market might react more sluggishly the farther we get from the moment the news gets out. And that the time slices are in human-friendly numbers understood by traders (and perhaps therefore affecting the reactions in the market: imagine some program set to trade 5 minutes after news), not the log scale beloved by mathematicians and scientists. Sometimes culturally appropriate numbers are more accurate than analytic ones (as well as more meaningful).

2) The range of movement in each time slice is drawn in full black, so that analysts can easily see both that range and how it relates to all the ranges before and after it. (Gray showed the cumulative range; perhaps confusingly.)

3) These black time ranges perceptually group into a stepped shape, designed so that when dozens of these shapes were drawn on the same page analysts could see which ones looked similar. If enough looked similar, they could say that the market reacted consistently.

Rearranging time-based ranges to allow algorithm validation

     

I developed this visual tool to help people determine whether adding risk-reducing rules to a “black box” trading algorithm would make it more profitable. The new rules being considered were a stop loss (“if the price gets this low sell immediately because we think it’ll go much lower”) and a profit goal (“if it gets this high that’s all we needed and we don’t think it’ll go much higher, so sell immediately”).

Horizontal bars are drawn in real dollars relative to the purchase price, so the index along the top shows exactly how much was made or lost on each share. They extend from the lowest to the highest (relative) price at which the stock traded while it was held by the algorithm. There’s a white hair line inside each bar that shows the price at the time the algorithm chose to sell. (This is barely visible in the full-size scan; click to see it. These were all designed for then-standard 300 dpi laser printers, and the hair line was tuned to be easily visible but not interfere with judgement of the black area, which was proportional to the total profit or loss taken). I drew a semicircular “cap” on the bar if the sale price was exactly at the top or bottom of the range.

In the first image I simply drew the trade ranges in time order—you can see something of a cyclical variation in the two and a half bulges reading from the top; in fact the data was two and a half years worth. The second rearranges the data by the amount of profit taken: see the hair line advance from losses to profits as you move down the page (again, visible only in the full-sized image). The third images shades the ranges by time: early trades are light, later ones dark. There is some indication that the darker trades group toward the bottom: this would lead to the testable hypothesis that the algorithm did better in the market conditions prevalent toward the end of the time span.

This “paper machine” worked in the following way: an analyst would move a pencil or ruler, held vertically across the bars, from the far right slowly towards the left and look at the amount of profit (black area) that would result if the algorithm used the ruler’s position as a profit goal. (We assumed we could sell immediately at that exact number, or at least that the deviations would be minimal or balanced.) If the amount of profit added by the profit goal (black area to the left of the ruler and the right of the hair line) was larger than the profit forgone (area to the right of the ruler and the left of the hair line)

This is work done with Hezzie Lamdan, and ironically it led to its own demise—a result I ironically consider a triumph of visualization. After seeing dozens of the marked and folded printouts around his desk for a few weeks, there were suddenly none. When I asked if I needed to adapt the tool to some change in the process, he said no; that after using it for a while he realized he could code this analysis technique into a meta-layer in the black box, and didn’t need the paper any more. Why is this a good thing? Because the visualization made it possible to understand the analysis technique better, by virtue of its clarity, and provided Hezzie the intellectual foundation that allowed him to do a deeper kind of analysis.

Blurring a scatter plot to find consistent areas

     

The goal of this visual manipulation was to make it easier to find areas where there were more positive points in a scatter plot (marked with a ‘+’ in the image) and others where there were more negative points (marked with ‘o’, since ‘-’ was too visually similar). We (Hezzie Lamdan and I) realized that simply blurring the image (“convolving each data point impulse with a Gaussian”) would address two issues simultaneously. It would reduce the visual noise—the main thing that was making comparing the positive and negative points difficult. (It was even harder before I got there, when just ‘+’ and ‘-’ were used.) And in the domain of the data, it would allow neighboring points, in a local area, to balance one another out: e.g., three small negatives might have been okay in this analysis if they were right next to a very large, more than offsetting, positive.

This is an interesting case where the visual solution (reduce noise) and the analytical need (offset neighboring data points) were solved by the same operation, played out in the two domains (blur vs. convolve). This is more than a lucky coincidence; it happened because the representation was a direct and accurate transcription of the data, as structured by the needs of the analytical task.

The first image is my re-drawn scatter plot, the second shows the same data convolved with a small-radius Gaussian, the third with a larger-radius Gaussian. Note how if I tell you that the white areas are the positive ones and the black negative, the third image makes identifying consistent areas trivial. (Click on the images to compare higher resolution versions.)