Results of the Fitness Challenge using SAP Lumira


During SAPPHIRE NOW in Orlando, the analytics team issued a “fitness challenge,” asking people to record their activity with a device such as a Fitbit or Jawbone Up and email us the results.

The resulting analysis was a microcosm of the kinds of issues that happen in organizations every day – along with announcing the winner, here’s an explanation of how I did the analysis using SAP Lumira (previously known as SAP Visual Intelligence).

The first thing I did was to take a look at the different files provided by the participants, in a variety of different formats (CSV, XLS). I could have used the powerful data merge capabilities of SAP Lumira (including the new “Union” option shown below), but for this one-off analysis, I used Excel to cut and paste the data into the right formats.


The first problem was the differences between the data sets generated by the Fitbit One and Jawbone One devices:

  • The Fitbit export columns were clearly labeled and easy to understand, whereas Jawbone ones were a little more cryptic. I eventually found an explanation of the different fields and was able to add some conversions to align the data (e.g. m_distance = distance in meters, so multiply by 0.000621371 to get miles)
  • The devices measure different things– for example, the Fitbit divides activity time into “lightly active,” “fairly active,” and “very active” time while the Jawbone takes a different approach, providing fields such as “longest active time” and “total workout time.”
  • The devices measured the same thing in different ways. For example, “calories burned” for the Jawbone seems to be just the calories burned during exercise, whereas the Fitbit estimates calorie burn for the whole day.

Once I had made all the adjustments in Excel, I created a new document, imported the data, and accepted the default “semantic enrichment” which correctly identified the measures and date field.



Then it was very easy to create a visualization of the number of steps taken by day (several people provided data for Monday, but I excluded it since it was a travel day for most people).

The results: the seven participants in the challenge took a total of 284,908 steps during the conference. As you can see from the chart, people started off with great intentions on day one, but couldn’t keep it up for days 2 and 3:


By breaking this result down by person and day, we can see that the the slowing-down trend was true of most people, except Daniel Graversen and Henrik Wagner (both Jawbone One users). Daniel, in particular, put in an all-out effort on day three to clinch the “most steps in a single day” title.


We can also see that the two Jawbone users, on average, were more active than the Fitbit users:


So who won the challenge? As ever, it depends (a bit) on how you choose to measure “winning!”

Daniel Graversen’s last-minute burst clearly wins him the bragging rights on the total number of steps (shown horizontally this time, just for show SAP Lumira does horizontal charts, too)


Looking at total distance, however, appeared to show a different winner – me! (Timo Elliott)


That looked a little strange, though, given how many fewer steps I’d taken. When I looked closer, I realized that I had another data quality problem I hadn’t seen in Excel: as a European, my Fitbit was exporting the data in kilometers, not miles, giving me a distinctly unfair advantage. This was easy to fix fix using the powerful data manipulation tools in SAP Lumira. I created a new column called “Distance in Miles” with the formula “if {Name}=”Timo Elliott” then {Distance}*0.621371192 else {Distance}” (the product provides hints at each step, so you don’t have to know the syntax).


The updated chart looked a little more realistic – it showed that Daniel Graversen also won in total distance, with over 30 miles during the conference. It also shows that Greg Myers beat me in distance despite my lead in steps – i.e. he’s much taller with a longer stride (this value can be set by the user under the advanced options…). Just for variety, this example uses color and a legend to indicate the names, rather than showing them on the X axis (this is not best practice).


There are, however, other measures that could be used to determine a “winner”! Jamie Oswald won in calories burned (even though the Jawbone numbers from Daniel Graversen and Henrik Wagner are probably not comparable, it doesn’t seem like they’d come out ahead of Jamie):


And, of course, we could drill into any of these measures to get the daily winners, eg Greg Myers had the highest number of steps on Monday and I had the highest number on Tuesday (this chart shows another of SAP Lumira’s many different available “palettes” for charting):


I also uploaded the same data to the (very) beta version of SAP Lumira Cloud, where I was able to do easy visualization of the prepared data set, and share it with others:


To conclude, I was fascinated to see that even this small-scale “quantified self” analysis was rich in the same kinds of issues that plague much larger and more important projects:

  • Multiple data sources in different formats (Excel files, CSV files, different column names for the same data, etc.)
  • Different data from different sources (the devices all measure effort and calories, but look at different aspects of the same activity)
  • Similar fields having very different definitions (“calories” for the Fitbit is for the full day, whereas it seems to be just for sport with the Jawbone)
  • Units confusion (is that distance column in miles or kilometers?)
  • Forgotten data (I initially overlooked Henrik Wagner’s data! –  the eagle-eyed among you will have noticed that it’s not in the SAP Lumira Cloud data set)
  • Manual copying and pasting of data (each person collected their own data and sent it by email, it took time and effort, I could have made errors during the cut-and-paste, etc.)
  • Possibility of people “tweaking” the numbers (this is just for bragging rights, but if there were monetary awards, it would be easy to game the system, e.g. by setting an overly-high stride length to get a high distance, or just shaking the fitness device to simulate steps!)
  • Multiple, competing KPIs that could be used to define success (what were we trying to achieve, anyway?)

If you haven’t already done so, you should give SAP Lumira a try. It is a great tool for collecting data, manipulating it on the fly, and sharing it with others (and until May 31st, 2013 you can get a copy of SAP Lumira Personal for only $9!).

Here’s a video version of some of the analysis above (sorry, without Henrik’s data!)

Many thanks to all the participants in the challenge, and I look forward to having more of you the next time we try this – maybe at the ASUG SAP BusinessObjects User Conference in September?