# Intraclass correlation coefficient: zero and negative

I am trying to calculate reliability between two raters for continuous data. The data is frequency of negative life events for each participant. My data looks like this:

Dataset 1

rater 1: 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0

rater 2: 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0

Dataset 2

rater 1: 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0

rater 2: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

This is continuous ratio data with a true zero (i.e. 0 indicates that there are no negative life events). The values can go beyond 1 but this is what we ended up with. My problem is I'm getting a negative ICC for dataset 1 (-0.205) and zero for dataset 2. How do I interpret this and report it in the write up?

• Welcome to CV. First of all, this doesn't look like "continuous ratio data" to me. Rather, it looks like ordinal, zero heavy count data. Next, based on these two datasets, there are no points of agreement in #1 and rater 2 in #2 has no life events at all. Finally, what is being rated? Is it a time series? Apr 1, 2016 at 11:12
• Sorry, I should have clarified what was being rated. Each point of data = number of negative life events per participant. You're right, it is count data. Raters decided if each event was negative or not and the values are how many were judged negative for each participant. It just ended up that many events were not negative, hence the zero heavy counts. Apr 1, 2016 at 11:22
• So, there are 16 participants being rated by two people for each dataset? How do the datasets differ? Are there 4 different raters? Apr 1, 2016 at 11:25
• I tried to simplify in presenting the data here and maybe have overdone it! They are actually from the same dataset (and same participants). Dataset 1 is for life events that are long term and dataset 2 is for life events that are short term. They are mutually exclusive in our data. Apr 1, 2016 at 11:29
• Forgot to add, its the same two raters in each dataset. Apr 1, 2016 at 11:30