# How to analyse individual consistency from three non-normally distributed variables?

I would like to know if individuals were consistent across three different days. I measured the latency of individuals for a maximum of ten minutes. Many individuals had a latency of 0 and some individuals never arrived and were therefore given a latency of 601.

The data on each of the three days is very non-normally distributed and I do not manage to transform the data (tried log, log10, sqrt etc). Running spearman rank correlations works on two of the three days but is not possible with three variables. Running ICC analyses is possible with three variables but does not seem valid with my non-normally distributed data.

How can I analyse the consistency of individuals across the three days? Could I rank the data and then run ICC analysis? Here is the dataset:

     ID d1  d2  d3
1   198 21  5   0
2   192 0   0   0
3   199 1   601 3
4   191 601 601 601
5   197 601 601 313
6   194 0   23  0
7   200 23  263 2
8   143 18  130 2
9   196 19  52  0
10  195 8   94  6
11  194 0   14  0
12  199 0   0   0
13  197 0   19  0
14  129 26  41  1
15  198 13  489 0
16  191 0   0   0
17  170 0   601 1
18  162 0   29  0
19  164 1   93  18
20  168 0   29  0
21  167 0   0   0
22  169 0   601 3
23  166 0   592 0
24  163 23  33  96
25  154 0   601 0
26  155 0   601 0
27  159 0   601 0
28  156 0   90  0
29  158 0   94  0
30  160 0   0   0
31  157 0   60  0
32  141 0   601 0
33  179 0   0   0
34  176 0   43  1
35  161 4   157 0
36  142 1   74  0
37  172 0   2   0
38  178 2   367 3
39  147 1   4   0
40  144 1   601 38
41  143 0   11  0
42  141 0   601 601
43  146 124 601 0
44  142 0   601 0
45  148 0   0   0
46  134 10  73  12
47  180 0   288 0
48  133 0   600 0
49  165 6   577 0
50  131 11  199 57
51  174 42  601 224
52  171 0   346 0
53  140 0   127 0
54  123 2   14  0
55  121 0   10  0
56  127 0   50  0
57  122 7   42  0
58  124 0   131 0
59  125 0   15  0

• What about Kendall's concordance coefficient? en.m.wikipedia.org/wiki/Kendall's_W Jul 11, 2014 at 10:06
• Thanks, that seems to work actually. I just want to make sure it really calculates what I want as this is behavioural data and not to compare the ratings of people. Also, how would you write up the stats of Kendall's W for a manuscript? Thanks Jul 16, 2014 at 13:10