Read on Twitter
I am posting a short thread about an analysis I have run on the heart-beat-counting task (HCT). The HCT has been heavily criticised and some of the issues raised are entirely valid. What follows is not an attempt to address these concerns 1/n
One concern is that it is not possible to generate a null space with which to compare the actual HCT scores. Here I have attempted to investigate this. I mainly want to know if there are any issues with the approach I have taken 2/n
I have managed to pull together actual and reported heartbeats from 1920 participants. The data is taken from a number of publicly available datasets (thanks @DesmedtOlivier , @JJMurphy90 @legrandni) and others that people have kindly sent me 3/n
the different studies use different window lengths so I have normalised the data to the actual number of HBs and the reported number of HBs in 60 seconds. 4/n
The first figure shows the distribution for the actual (blue) and the reported (red) counts/minute. The main thing to note is that the reported HB is lower than the actual HB counts. This is not new and others have reported this previously 5/n
I then calculated the HCT score using 1-(abs(actualHB-reportedHB)/actualHB). I then calculated the distrubution of these scores. To create a null space I randomly shuffled the actual and reported HBs and recalculated the HCT score 6/n.
This was repeated 100,000 times and the distribution of shuffled the HCT scores was calculated. From this I calculated the 5 and 95th percentiles to see if the actual HCT scores differed from this distribution 7/n
The results are shown in the following figure. The black lines are the shuffled data with the percentiles plotted. The blue green and red lines show the actual HCT. The green shows data points greater that the 95th percentile and red lower than the 5th percentile. 8/n
of interest is that there are points at which the actual HCT scores differ from signifcantly different from the shuffled data. Of particular interest are the values close to an HCT of 1. 9/n
To check that this was not some issue with taking the absolute value I repeated the analysis calculating the HCT score as 1-(actualHB-reportedHB)/actualHB. 10/n
This showed the same result. HCT scores between 0.875 and 0.95 occurred more than one would expect by chance from the shuffled data. Also the symmetric scores greater than 1 occurred less than chance. 11/n
This would seem to suggest that there is an increased frequency in reported HBs that are close to the Actual number of HBs but that are under reported. 12/n
To check this I repeated the analysis simply looking at actualHB-reportedHBs. Now it is really clear that there is an increase in proportion of people close to but under reporting the number of HBs and a decrease who over report. 13/n
This is of interest as it has been proposed the HCT score should be used when the HCT is higher than 0.85. In other words there are & #39;good perceivers& #39;. This analysis would seem to support this. 14/n
However, before worrying about the interpretation I am keen to see if I have done anything fundamentally incorrect. I am not 100% sure how to interpret the results and I also happy to discuss this. Thanks end/n
@manos_tsakiris @DrSFink @opatcorneille @micahgallen I am interested in your thoughts. Particularly if you think the approach is invalid! Thanks
