Weartime detection changes from GGIR 1.0 to 2.0 versions

[marcuslop...@gmail.com]

Hi Vincent and colleagues,

I hope everything is fine with you in these challenging times.

I had previously run GGIR 1.9-2 on my research data. Recently, with the release of version 2.0.1, I repeated the analysis on the same data and using the same script. The code is attached below.

The proportion of participants with valid data reduced by almost 10%. I compared daily estimates of part2 and part5 files trying to identify the reason behind this. The image below illustrates the comparison of wear time estimates of the same participant who wore the accelerometer attached to the wrist for six complete days (24h). I understand that part5 of GGIR 2.0.1 does not analyze days with less than 23h, which explains the lines with no data. However, two more days were excluded from part5 analysis when comparing GGIR 2.0.1 and 1.9.2 outputs (highlighted lines). 

I read the GGIR update notes from 1.9.2 to 2.0.1, but it is not clear to me which change may explain the loss of valid days.  I appreciate any help you can give.

[marcuslop...@gmail.com]

I am sorry for the bad quality of the image.

The pdf version is attached below.

[Vincent van Hees]

Hi Marcus,

Part 4 and 5 of GGIR have undergone major improvements this year, thanks to both contributions from a number of people. The code changes we make are always intended to improve the code, but bugs can occasionally slip in which are then later resolved.

Key changes are listed here https://cran.r-project.org/web/packages/GGIR/news.html, which includes a number of changes that potentially explain your changes in valid days. Additionally if you are familiar with GitHub you can compare the code between specific versions of GGIR with GitHub's code compare functionality.

If you identify mistakes in the latest version then post them in the GitHub issue tracker: https://github.com/wadpac/GGIR/issues. GGIR is open source. The more people actively monitor consistency and dedicate time to investigate potential problems the better the GGIR will get.

Best,

Vincent

Dr. Vincent van Hees

Independent consultant

https://accelting.com/

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐

--

You received this message because you are subscribed to the Google Groups "R package GGIR" group.

To unsubscribe from this group and stop receiving emails from it, send an email to RpackageGGIR...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/RpackageGGIR/7ff8b625-988a-40ce-8481-34f70fe1c235n%40googlegroups.com.

[Vincent van Hees]

The changes in your pdf can be explained by the fact that we are now intentionally excluding first and/or last day of the recording when they are incomplete. This is because it would otherwise distort the person level time use summaries. In order to have meaningful compositions of time-use at a person level it is critical that day level summaries are complete.

Vincent

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐

--

You received this message because you are subscribed to the Google Groups "R package GGIR" group.

To unsubscribe from this group and stop receiving emails from it, send an email to RpackageGGIR...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/RpackageGGIR/cf455889-156c-4220-b4fa-3bc63c051c52n%40googlegroups.com.

[marcuslop...@gmail.com]

I observed that most of the changes in part4 and part5 outputs of my data are related to the exclusion of the first and last days.

But I am trying to understand which improvement in GGIR may have affected a significant loss of valid days in part2. In this same PDF, we can see two days (10/13 and 10/14) with 24 hours of valid data in GGIR 1.9.2 version, but with 9 and 4.25 hours in GGIR 2.0.1. It is a huge difference that occured in the middle of the seven days protocol.

I checked the github tracker but it is hard for me to understand the code.Do you have any clue of what could provide such a loss of wear time in part2?

Thank you for your support!

[Luís Eduardo Argenta Malheiros]

Dear Vincent,

We observed similar differences in the estimates of behaviors between versions as described by Marcus. Could you briefly describe the algorithm updates that may explain the changes on part2 estimates of non-wear time?

[Vincent van Hees]

Dear Luis,

See my response to Marcus on the 17th of October. GGIR is open source software. We do our best to improve the code and communicate changes as good as possible, but ultimately open source software comes without warrantees.

GGIR is a complex software with hundreds of potential use cases. We have various measures in place to monitor functional consistency, but just like any other complex software project it is hard to make this 100% error proof. The best solution is to have a critical and pro-active user community to spot and fix issues quickly.

Researching the cause behind specific changes in output for specific files between specific versions of GGIR is time consuming. I am happy to do this for you as a paid consultancy. GGIR may be for free, but my time is not.

Thanks, Vincent

Dr. Vincent van Hees

Independent consultant

https://accelting.com/

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐

To view this discussion on the web, visit https://groups.google.com/d/msgid/RpackageGGIR/e7c8e6af-fa52-4a71-b5d6-0362fe113714n%40googlegroups.com.