HackMotion is an incredibly useful piece of tech, for both coaches and players. But how accurate is it? And how can you get the most accurate data when coaching and practising?
We did the testing to save you time, I know most won’t want to analyse the full dataset, so below are the key finding and top tips for coaching and practice, followed by the full dataset. You can also check out this link for our broader HackMotion Golf review.
Table of Contents
Key findings
The key findings are as follows:
- HackMotion is reliable within 0-3º.
- HackMotion is valid within 0-5º for most ranges*.
- The trend is for HackMotion to slightly underestimate max angles.
*Calibration is super important for valid values.
These conclusions are based on our testing data covered later within this user guide.
User guide for coaching and practice
Reliability vs validity
Reliability and validity both refer to how well a tool measures something – reliability refers to how repeatable a measure is. Whereas, validity refers to the accuracy of the measure.
If a player had a wrist angle at impact of º10, 10º, 10º over three swings and Hackmotion measured 25º, 25º, 25º we would say it was not very valid, but highly reliable. If Hackmotion measured 21º, 25º, 29º we would consider it more valid, but less reliable.
From this example, you may start to see that both are important constructs depending on how you plan to use HackMotion.
HackMotion is reliable within 0-3º
Across all ranges of movement, HackMotion is highly reliable (0-3º). Variance within 3º might be measurement error, but if you see changes over this value within a lesson/practice you can be confident they reflect a real change in the movement.
For me, as a coach, this is the most important takeaway message.
HackMotion’s validity is 0-5º when correctly calibrated
We ran the entire testing twice (three times if you count pilot testing). The first time all data were within 5º apart from flexion. The second time all data were within 5º apart from radial deviation.
This is down to the angles at calibration and how HackMotion ‘zeros’ values. As a coach, I’m not too concerned with this, as long as the data is reliable. Apart from any calibration error, HackMotion showed a trend to underestimate by a degree or two, rather than overestimate max values.
How to get super-valid measures
During our testing, we followed HackMotions guidelines for calibrating once on the wrist (our mechanical wrist). However, if you want super valid measures (35º to read 35º) we’d suggest calibrating the sensor on a flat table, with the sensors vertically aligned and controlled in both positions as shown below.
This approach allows all ranges to start at zero and account for any displacement from this position.
Make sure the sensors are placed directly onto the player’s arms without the two units rotating. Remember validity isn’t the only important factor, but if this is what you care about it will help.
Player progression
Where coaches need to be careful is when they are re-testing after a few days or weeks. Any difference in the positioning of the sensors or calibration process will result in different numbers (not less reliable data).
If you are bullish on tracking numbers, calibrate as outlined above, then draw around the sensor placement after calibration on the player’s arm, take a picture of where the sensors were placed and use this to best replicate the calibration and testing on the next session.
A better option is to just focus on trends and graph shapes, not raw numbers between lessons. Be aware that a 10º jump between lessons could just be your own calibration or sensor placement error (welcome to biomechanics – yes it is boring).
Crosstalk
Watch the video below, particularly the last section where the joint is locked and both sensors are rotated. You’ll notice the other ranges fluctuate by a few degrees – this is called crosstalk and is present in all devices of this nature.
In the final testing, we even used lasers to check we weren’t deviating off plane during our testing. There is some cross-talk present with HackMotion, but it still works well. There is nothing you can do, other than be aware that it is present.
Crosstalk is a more serious issue when one sensor rotates independently of the other (not recorded but try it yourself), both flexion/extension and radial/ulnar deviation numbers will change drastically and sometimes get stuck in extreme rotations (gimbal lock).
When coaching, make sure the sensors are firmly strapped to your player, independent rotation of the sensors will make your measures both less valid and reliable.
Players with arms that are too small to secure the devices, players with more fat mass and higher arm speeds will all be cases where the sensors may twist/move on the skin (we call this skin movement artefact).
The HackMotion straps are really good, but feel free to add extra strapping and tape if you have some funky data and you suspect one of the above.
That wraps up our user guide – I hope it is of use and helps you improve your own golf and the golf of others. If you want to explore the full dataset keep reading.
HackMotion testing video
Below is a sped-up testing video showing how we collected with a 2D camera and the HackMotion app. This is a very low-tech, but surprisingly accurate method if the perspective error is minimised and you can accurately locate the joint centre during the post hoc analysis. The order was as follows: calibration, flexion/extension x 9, radial/ulnar deviation x 9 and rotations x 9.
HackMotion sensor data
The following table shows the data present in the HackMotion app for every measured max position.
| Max Flexion | Max Extension | Max Radial | Max Ulnar | Flex/Ext Range | Radial/Ulnar Range | |
| 1 | -34 | 37 | 31 | -37 | 71 | 68 |
| 2 | -35 | 38 | 32 | -38 | 73 | 70 |
| 3 | -35 | 40 | 30 | -38 | 75 | 68 |
| 4 | -34 | 38 | 32 | -34 | 72 | 66 |
| 5 | -35 | 39 | 32 | -38 | 74 | 70 |
| 6 | -38 | 41 | 33 | -32 | 79 | 65 |
| 7 | -38 | 38 | 31 | -39 | 76 | 70 |
| 8 | -38 | 37 | 29 | -41 | 75 | 70 |
| 9 | -38 | 35 | 27 | -41 | 73 | 68 |
| Average | -36.1 | 38.1 | 30.8 | -37.6 | 74.2 | 68.3 |
| SD | 1.83 | 1.76 | 1.86 | 2.96 | 2.39 | 1.87 |
2D video data
The next table shows the measured 2D angle in the max position.
| Max Flexion | Max Extension | Max Radial | Max Ulnar | Flex/Ext Range | Radial/Ulnar Range | |
| 1 | -37 | 40 | 43 | -35 | 77 | 78 |
| 2 | -37 | 41 | 41 | -32 | 78 | 73 |
| 3 | -34 | 40 | 43 | -33 | 74 | 76 |
| 4 | -34 | 41 | 45 | -34 | 75 | 79 |
| 5 | -36 | 42 | 44 | -34 | 78 | 78 |
| 6 | -39 | 40 | 43 | -31 | 79 | 74 |
| 7 | -35 | 41 | 43 | -34 | 76 | 77 |
| 8 | -37 | 42 | 41 | -34 | 79 | 75 |
| 9 | -38 | 41 | 44 | -34 | 79 | 78 |
| Average | -36.3 | 40.9 | 43.0 | -33.4 | 77.2 | 76.4 |
| SD | 1.73 | 0.78 | 1.32 | 1.24 | 1.86 | 2.07 |
You can see all data are a close match apart from radial deviation.
Difference by trial between HackMotion & 2D video analysis
The final table shows the difference between to two measures for each trial. The level of consistency in each column shows HackMotion is highly reliable.
| Max Flexion | Max Extension | Max Radial Dev. | Max Ulnar Dev. | |
| 1 | -3 | 3 | 12 | 2 |
| 2 | -2 | 3 | 9 | 6 |
| 3 | 1 | 0 | 13 | 5 |
| 4 | 0 | 3 | 13 | 0 |
| 5 | -1 | 3 | 12 | 4 |
| 6 | -1 | -1 | 10 | 1 |
| 7 | 3 | 3 | 12 | 5 |
| 8 | 1 | 5 | 12 | 7 |
| 9 | 0 | 6 | 17 | 7 |
| Average | -0.2 | 2.8 | 12.2 | 4.1 |
| SD | 1.79 | 2.17 | 2.22 | 2.57 |
Summary
Accurately testing validity and reliability is challenging (previously published work here and here). And creating a device that accurately measures movement, like HackMotion is incredibly tough. This isn’t an academic-level analysis, but I hope it allows players and coaches to understand how accurate the numbers are and some pitfalls to look out for.
Feel free to leave any comments below and I will get back to you.
Happy golfing – Will @ Golf Insider
