Wednesday, July 11, 2012

input please: competing visualizations

As part of my day job, I consult internally on data visualization. I was recently working on some mock visualizations for a project being undertaken by my team. Since I can't talk about the details here, I've created another scenario to use to explain some generalized versions of the visuals and show the iteration process that I went through to get to what I thought was clearly the best visual.

Challenge: of the four versions of the visual I put together, the project manager favors the one I like the least.

So I thought I'd try to leverage the wisdom of the crowd (you, in this case) and see where that takes us. I won't bias you by telling you which visual the PM favors and which I do (though if you've been reading for a while or have attended one of my workshops, it will likely be clear). Rather, I'll simply describe a scenario and then show you the made up data and the four mock ups that I put together. Then I'll let you be the judge.

What I'd like to know from you is: which is your favorite and why?

Here's the scenario*: you are going to administer a personality test and your challenge is to visualize the results to give back to the individual in a way that is straightforward to understand. There are five different personality profiles, each of which has four measures that are expressed on a scale with possible values 0, low, medium, or high.
*Note that this scenario is totally made up (to protect some confidential stuff), but should provide you the sort of context you'll need to evaluate the effectiveness of the visuals.

Here is the made up data that I used for my mock ups:


Here are the four mock up visuals I put together based on the above data:








Which do you like best? Why? Leave a comment with your thoughts!

28 comments:

  1. Probably would've gone w/ frequency polygons or sparklines, personally. But of those, I tend to prefer 3. Easiest to put an actual measurement alongside without being overly complicated like 4.

    ReplyDelete
  2. slightly unclear from the context - are the 4 measures across the 5 different profiles comparable with each other?

    if yes, then i would vote for the third mockup, as it seems to convey this in the best manner. if not, then i favor the 4th mockup because of its use of bar graphs (which seem easier to gauge at a glance than gradient changes).

    ReplyDelete
  3. I think you've anonymized away too much information! I'm not sure whether each person taking the test receives all five profiles (e.g. "Intelligence Profile", "Sociability Profile", etc...), or if each person receives the most appropriate single profile (e.g. "Introvert", "Extrovert", ...).

    I assume the former, since that's what the visualizations seem to indicate, but it's hard to tell from the description.

    It's also unclear whether the measures are "similar" (in which case the last might be appropriate) or "different" (in which case the last is probably inappropriate).

    All that said, I would say:

    #1: I don't like the weird spatial arrangement unless there's some relevance to it that's obscured by the anonymity.

    #3 is worse than #2 because in #3 it's easier to compare the same measure across profiles. (Assuming you'd want to do so.)

    #4 is probably best if the measures are numeric and all using the same scales (say, 1 to 10, or 0% to 100%). If they're not then I would find it misleading and I'd prefer #3.

    ReplyDelete
  4. I prefer mock up 3, it enables comparisons within both profile rows and measure columns, so I can easily see most left measure 3 blank, profile 5 was the highest, and profile 4 was the lowest.

    mock up 4 could also be interesting, along with switching the hierarchy as another option.

    See http://public.tableausoftware.com/views/mockup/Sheet2 for a possible combination of the two (with a couple of other sheets as tabs).

    ReplyDelete
  5. Hi everyone - thanks for the comments! A couple clarifying points :

    1. yes, each person will see their results across all 5 profiles (it's likely scores will be highest for the measures across a single profile - e.g. profile 5 in the example - though that may not always be the case.

    2. The four measures across the profiles are NOT the same (sorry, this was confusingfrom my naming), rather each profile has four measures that are distinct from the measures within the other profiles. So while all are on the same range, you wouldn't be trying to compare the 2nd measure in profile 1 to the 2nd measure in profile 2, etc.

    ReplyDelete
  6. If the measures aren't supposed to be comparable across profiles, I'd consider a different color for each profile. And I'd change my answer to #2 or #4, since #3 encourages me to compare measures between profiles.

    Alternatively, if you wanted to stick with #3, I'd think about sorting the measures within each profile from highest to lowest rather than in some fixed order.

    ReplyDelete
  7. Mockup 1 puts the 3rd profile in the center, giving it special attention/importance.

    In profile 2, I got slightly confused by the legend at the top, which suggests (to me) that the horizontal placement will be like that. So I expected profile5-measure1 to be a block in the rightmost column (vertically aligned with the legend 'high', which is empty now). I'd try to merge mockup 2 and 5 so you have blocks like in 2, but horizontal indicators on whether a value is high or low.
    The profile5 would have an empty 'left column', 1 block in the middle column', and three on the right. When I'd see that, I'd probably change my mind again.

    ReplyDelete
  8. What's the goal of conveying this information? What do you want them to do when they see it? Is it to get them a blended score (overall you're slightly below "mid"), or is it to get them to focus on certain elements, either the low or the high, or perhaps the ones deviating from the norm?

    The way you have it colored now on #1-3, my eye is drawn to the ones that are "high". Is that what you would want me to look at? If not, then I would suggest finding a way to color #3 or #4 to call out what you want, or rearranging one of those to highlight the appropriate elements.

    ReplyDelete
  9. I like mockup 4, because I do not have to refer to a color chart to determine the value of each measure.

    ReplyDelete
  10. I prefer # 4 because of the alignment of the measures to the scale, but a score of score of 0 could be perceived as missing data. I also like #2 because of the organization and color gradation, but the size and placement of the scale throw me off a bit.

    ReplyDelete
  11. I found Mockup 4 to be the easiest to understand at a glance.
    The scale is directly in alignment with the scores and there is no need to do extra interpretation (i.e. translating a color to value.) as with the other graphs. I found this translation difficult in cases like mockup 2 profile 5 when only 2 colors where involved.

    ReplyDelete
  12. #4. Frankly, initially I didnt spend longer than 5 seconds on each of the four charts. The initial impact of the fourth chart is the most powerful in its clarity and interpretation of groupings. For what it's worth, the preceeding three were cluttered in a way that i couldn't (and didnt want to read)

    ReplyDelete
  13. I'd say #4 is the easiest to read, allowing easy comparison across the profiles and measures without having to check back to the colour chart.

    ReplyDelete
  14. I like 3. It is visually engaging and only 4 graduations of the hue make it easy to interpret.

    ReplyDelete
  15. I like #4 - it's the most clear, and presents the information in a way that allows you to quickly gauge your results.

    ReplyDelete
  16. i like 4# too ... it is only one where i understand something
    best regards
    r

    ReplyDelete
  17. I like #4 - it's a clear point of seeing quickly the low to high measurements easily for each profile. It seems this would be the goal for an eye catching assessment.
    Thanks!

    ReplyDelete
  18. I find #3 to be the simplest and therefore the one that communicates the best. The color hues communicate your findings clearly and are visually engaging. #4 is a little too complicated and, more importantly, mis-communicates the findings. By putting the Low/Medium/High labels (which indicate ordinal level findings)on a horizontal scale it suggests interval or ratio level differences in the findings that are not really there.

    ReplyDelete
  19. #4 was the easiest/quickest for me to establish what the baseline is, as opposed to using color.

    But I think I would have arranged the profiles from highest to lowest and then ranked each measure highest to lowest within each profile.

    ReplyDelete
  20. I vote for #4. But I'm guessing I'll learn more today about which you prefer vs. the PM. :) Tell us the result!

    ReplyDelete
  21. #3 without the borders please cole

    ReplyDelete
  22. #3. It's the most scan-able!

    ReplyDelete
  23. I'd suggest an approach sort of starting with what you have in #3. A regular table, first column is profiles, and each remaining column header for each of the 4 metrics to be some cute graphical representation of what the measure is.. e.g. Smile quality, Friendliness, Smartness, whatever. And perhaps also an aggregate column that does a weighted average of the 4 metrics. Each of the column headers can be clicked to help you sort the profiles by that metric. You can still color code each cell based on the value there. but this approach is less verbose.. doesn't repeat the measure a few times like #4. And provides an interaction I think the user would appreciate.

    ReplyDelete
  24. #4 was the easiest to understand quickly. #3 could work if the color legend was flushed right to line up with the measure items - the way it's position threw me off - however, the only reason I was able to understand #3 was AFTER I saw #4... :-/

    ReplyDelete
  25. #4, absolutely. To do what you do I'm sure you are very fond of making data visually exciting, but most people really aren't. One of the few lessons that 'stuck' for me in web design was not to make the viewer (customer) have to figure out your clever navigation. Most people only want to know; do you have it, what does it cost, how can I buy it and how do I get out of here?

    ReplyDelete
  26. #4 is immediately consumable. Virtually no mental overhead. Simple and powerful.

    ReplyDelete
  27. I also vote for #4 -- by far it is the easiest to understand. In the other three, the horizontal color key above the data makes it very confusing, because you are trying to vertically align that with the data that appears below. -- Nicole

    ReplyDelete
  28. I too find #4 misleading because of the level of measurement. I favor number 3 but agree with another poster that using the same colour scheme across all 5 domains might encourage comparisons between domains. I would work on the size and placement of the legend however.

    ReplyDelete