A Year with the Fiat 500e

While this has nothing to do with fitness data it is data oriented. My Fiat electric car has given me a new world of data to play with, and as I complete my first year with the 500e, I wanted to share some of the data I have been collecting.

What I learned looking at the data

    • The percent charge dashboard indicator is very accurate.
    • On average the MPGe indicated on the dash is 17% too high.
    • On 1 hour of charging you can drive:
      • 3.8 miles if charging at 120V
      • 12.5 miles if charging from a 240V TurboCord
      • 20 miles if charging from a Level 2 Charger

Details

Each time I charged the car I collected the percentage of charge remaining, the miles driven and the MPGe estimate. When I charged at public stations that provide statistics on the number of kWh charged I tracked that also. The raw data is here.

In trying to look at the accuracy of the charge indicator I needed a known reference to measure against. The only way I found was to use the measurements of ChargePoint charging stations. Whenever I charge the car at a ChargePoint station it will give me an exact reading of the kWh used.  I then compare that to the change in the battery charge indicator.  The computed kWh is found by:

(ending percentage – starting percentage) * 24 kWh (the battery capacity)

Using this I plot the computed kWh against the measured kWh.  Doing this produces:

fig_1

As you can see the values match very well.

Once I established the accuracy of computing the kWh consumption based on the change in the percentage of the battery charge indicator I could use this to understand the accuracy of the other dash readouts.

Comparing the MPGe (or miles/kWh) estimated on the dash to the values computed, I found that it was the results varies quite a bit. As we can see here the computed MPGe for a given dashboard MPGe, have a wide range.

fig_2

If we look at the error in each data point and plot that on a histogram we get:

fig_3

From which I computed the average overestimation of 17%. Even though the dashboard indicator over-estimates the MPGe my actual MPGe after a year of driving was 115 which is higher than the value quoted by Fiat.

When looking at EVSE’s (chargers) they specify the amount of time it takes to fully charge a car.  I don’t find this that useful. I often worry that my current charge is not adequate for where I’m trying to go, and I want to know how long it will take to charge to get X extra miles. I found that expressing charge rates in miles of driving per hour of charging to be useful for this. The results above are averages based on 66 partial charges. I did tend to notice that the car will charge slower as it approaches 100% so you can actually get above these rates if you are only partially charging the car.

Advertisements

Fitbit vs Apple Watch for Step Tracking

When I picked up an Apple Watch, I had already been carrying a Fitbit for two years. It was a habit I was reluctant to change. I was unsure how well the Apple Watch would work as well as how I would get data off the watch. At launch, there appeared to be no way to get the data without writing your own iPhone app which is not a path I wanted to go down. My solution was to wear both devices, a practice that I continue with today.

Over time, and the release of a new watch operating system, more apps appeared which can access iHealth data. Now that there is a way to get data from both devices, I decided it would be interesting to take a look at how the data compare.

Data Sources

My Fitbit data comes initially from a Fitbit Flex and since November 22, 2015 from a Fitbit One (after the Flex died). For the Apple data, the data prior to June 1, 2015 came from the sensors in the iPhone 6 (which can be used as a step tracker on its own) while the data after that date is from the Apple Watch. Because the iPhone and Watch are such different devices, especially in how they are worn, I have separated these into 2 different data sets and in this analysis I will only look at the Watch data.

Step Analysis

Steps are the base data that all devices record so it seems natural to plot the steps from the devices against each other.

Screen Shot 2016-05-08 at 8.37.17 PM.png

There is a clear linear relationship in the values reported by both devices but the variance is more than I expected. In this format it is very difficult to quantify this difference. A better way of understanding this is to compute a percent difference between the two step values and look at these in a histogram. While we’re at it we can also look at the differences in the distance data as well.

Screen Shot 2016-05-08 at 8.41.08 PM.png

This gives us a better feel for the differences. Much of the variation in the steps is below 20% and while the delta occurs on both sides, the tendency is for the Watch to read a higher step count than the Fitbit. We also see that the difference in the steps tend to be smaller than the differences in the distance which we will investigate later.

While this histogram gives me some level of comfort that both devices give acceptable results the one thing this does not show is whether there are accumulated differences over time. This we can see by plotting the cumulative steps over time.

Screen Shot 2016-05-08 at 9.23.33 PM.png

Looking at the data this way gives me comfort that both devices do an equally good job of tracking steps. There are times when I wear my Watch while not wearing my Fitbit (usually while cycling) and I have no trouble believing that the actually difference between the two would be lower than the 1.7% shown here if I could account for that.

Distance Analysis

Now that we have seen how well steps match, let’s look at the distance computed by these devices. Since the Fitbit has no way to measure distance, it has to perform a conversion from steps to distance. The Watch and iPhone combination have significantly more sensor data to work with, including a GPS, but I don’t know how these are used to determine the distance it logs.

Let’s look at the steps-per-mile that result from dividing the steps by the distance in miles generated by both devices.

Screen Shot 2016-05-08 at 9.26.49 PM.png

This gives us insight into how distance is handled by the devices. Whereas the distance produced by the Fitbit has a very simple relationship to the steps, in the case of the Watch the number of steps-per-mile varies significantly.

If the Watch is trying to measure the distance using the GPS I could see 2 sources for the variation in steps-per-mile. The first is just stride differences due to walking speed. A larger possible source of error could be how the Watch determines whether you are walking or doing some other activity. Since I wear the watch cycling I wanted to see if the steps-per-mile variation was different on days I wore the Watch cycling from those where I did not.

Screen Shot 2016-05-08 at 9.33.58 PM.png

As I suspected, the variation of steps-per-mile is much larger on cycling days. If the goal is to get an accurate walking distance from an Apple Watch, it should not be worn while cycling.

Conclusions

Through this exercise I’ve gained confidence that I could replace my Fitbit with my Watch for step data collection. The path to get data is a solved problem and the step data matches the Fitbit quite well.

I consider the distance differences a toss-up. Getting distance from a the simple computation of multiplying steps by a constant factor is at best an approximation. For walkers, this may be acceptable but for runners this won’t work as their walking and running strides could be quite different. If you plan to run, as well as walk, I would expect the Watch to produce more accurate distance results. A Fitbit model with built-in GPS would also be an option.

Source Material

For those interested, this post is provided as a Jupyter Notebook.

 

 

 

 

Did I Meet My Goal

Like many people that have fitness trackers, I set a daily goal. It’s great to look at the end of the day and see if I have met my goal but when I want to look back at a month, or a year, to see how well I’m doing it is pretty hard to do.

I can go to the fitbit.com site and look at a month of data and see

Screen Shot 2016-05-06 at 1.16.43 PM.png

but it is hard to see which days I did not meet my goal from this. If I want to look at a whole year, all I can do is look at monthly totals which is not very useful.

I knew I needed a pretty compact format if I wanted to see a whole year at a time. Something like a scatter plot.  Then I remembered a chart I had see at a Tufte course and knew what I wanted.  By converting my days from steps to either “goal met” or “goal not met” I could simplify the presentation.  Using a skinny bar chart I arrived at this.

Screen Shot 2016-05-06 at 1.23.14 PM.png

This allows me to see a full year with the strokes going up being days that I met my goal and strokes going down being the days that I missed my goal. This make it clear that I’m missing my goal way too often but also that the days I miss my goal are not as isolated as I hoped with some long streaks.

Another benefit of this representation is that I have one more dimension to play with: color. I speculated that my out-of-town travel was a reason I was missing my goal. By using red as the color when I’m on work travel, blue for vacation travel and green for when I’m home I could turn the image into this.

Screen Shot 2016-05-06 at 1.28.33 PM.png

Here I can see that when I’m on vacation I still do pretty well but there are a lot of those goal misses on work travel days.

Another perk of this format is that it is easy to stack these up to show multiple year at once. Here is 2 full years of data.

Screen Shot 2016-05-06 at 1.30.15 PM.png

I’ve found this presentation useful. I can now imagine using color to represent which days I went cycling or days when it rained. Lots of options.

Let me know if you have found a good way to visually represent step data.

 

 

Getting fitness data from an Apple Watch

When I got an Apple Watch last year I knew it would not work with the Fitbit platform easily (the two companies don’t like each other much).  I like some aspects of the Apple Watch but the fact that you can only look at the data on your phone or tablet makes it a poor competitor to the Fitbit platform.  There are apps that will sync your Apple Watch steps with your Fitbit account if you’re looking to replace your Fitbit with the Watch but I wanted to collect data from both so that I could compare them.

Now that I’ve had the watch for almost a year I wanted to get my step data from it and see how it compares with the Fitbit data.

The first path I found from a web search was to export the data from the Apple Health app. While the procedure works, it is painfully slow. For me it took over 20 minutes and my phone could no be used during the process. When I finally got the data it turned out to be a 60Mb XML file and I had to find another tool to convert that into a file that could be read into Excel. It just seemed there had to be a better way.

More web searching and I happened upon a free app named ‘QS Access‘ which allows you to select which Apple Health data you want and creates an Excel file of that data.  The app is quite fast and solved my need perfectly.

From a quick look at the step and distance data the first thing I noticed is that since a phone knows the distance it covers you get actual step and distance data without needing to specify a stride as you do with Fitbit.

I’m looking forward to seeing how the Fitbit and Apple Watch data compare but that is for another time.

Gender bias in Electric Cars

I did a survey of 215 electric cars in the San Jose, CA area from May 11 to June 3, 2015, taking note of the gender of each driver.  The survey covered not only traditional commute times but also midday times and weekends.  The results indicated that 64% of the electric car drivers were men.

I wanted to get a feel for whether this could just be measurement bias.  I determined that one, imperfect, way to check for this was to select a popular car model and perform the same type of survey. I logged 389 Prius drivers from May 21 to May 30 and found that 49.9% were driven by men.

This leads me to believe that the gender bias in electric cars is real.  This could be due to many factors such as the technology still being relatively new or range anxiety.

Update: 2016-06-25
To get a feel as to whether the maturing of the electric car market would affect my results I performed a new survey using the same methodology from May 24 to June 3, 2016. In this sample of 277 vehicles I found that 50.9% were driven by men.  I find it interesting that there was such a large shift in such a short time.

Thoughts on Microsoft Band

My introduction to fitness trackers was the original Nike Fuel Band which I really liked but after a year I switched to a Fitbit Flex because the Fitbit website provides better analysis capabilities. I’ve missed being able to use my tracker as a watch which is a limitation of the Flex. I’ve always wanted to track heart rate also.

When I first saw the Microsoft Band it appeared to solve both these issues and I knew I had to try one. I used it for 2 weeks. I saw a review of it done by Katherine Boehret and completely agree with but I think in terms of bullets rather than long form reviews so here are my thoughts from a Fitbit user’s perpective.

The Good

  • The color display is high contrast and easy to read.
  • The touch screen makes navigation easy
  • The heart rate function works great and seems accurate
  • The sleep function tracks sleep and resting heart rate which is great
  • Having vibrating alerts on my wrist instead of on my phone was fantastic

The Bad

  • Wow is this thing uncomfortable. It is not stiffer than my Nike Fuel but infinitely less comfortable. It was not just uncomfortable when sleeping but I often had to take it off when typing.
  • To get heart rate you have to wear it pretty tight so it does not move around like a watch which adds to the discomfort.
  • The battery life is an issue. If you use it to track exercise you have to charge it everyday and if you use it for sleep tracking that can’t be at night.  Since you also want to track steps it’s hard to charge during the day. Makes it hard to find a good charging routine.
  • Access to the data is super-basic.  You can only look at the data on your phone (no web interface) and it has the most limited stats. For a device with so many sensors the ability to analyze data is pathetic.

In the end I gave up on it but it will have a lasting effect on me.  I was an original backer of the Pebble smart watch which really soured me on smart watches due to it’s poor user interface and lack of functionality. The Microsoft Band really changed my perception of what a smart watch could do and I find it infinitely more useful than the Pebble. And it is fun to use.

Hopefully some of the new defines coming out from Apple and Fitbit will give me the capabilities of the Microsoft Band with better ergonimics.

Consolidating My Data – EveryMove.org

My first step towards taking control of my fitness data is to see if someone has already solved this problem. With a little research I came up with a few interesting possible solutions. One of these is EveryMove.org

EveryMove wants to “Track every active move you make, from archery to Zumba, using the apps and devices you already know-we support hundreds.” This sounded quite promising and I was excited to try it.

The Good

The signup process was easy and I found that it really does support a lot of devices and it was easy to connect to my Fitbit and Garmin accounts. This was really encouraging. It seems like a neat system for aggregating all your activity data into one place.

The Bad

Unfortunately it did not take long to find that this service was not what I was looking for. The focus for this site is to encourage you to be more active by letting you win badges – a great goal but not what I need. The site has no capabilities to analyze or even property review your past activity data. Nor did it try to pull in any of my old activity data, only reading in 3 days of my Fitbit data. It is not trying to be a repository for your data long term.

Conclusion

While EveryMove does not solve my problem it was encouraging to see that they could read data from so many devices.  This means that it is possible to aggregate all my data in one place, at least technically.

One Last Note

In looking into EveryMove I found that they have a pretty comprehensive site comparing activity trackers which is well worth looking at.