Reduce the range of time in sequence analysis with R - r

Question

I have a sequence that happens over a very long period of time. I tried 8 different algorithms to classify my sequences (OM, CHi2,...). Time goes from 1 to 123. I have 110 individual and 8 events.
My results are not as expected. First, it's very difficult to read. Second, a category contains too many representatives sequence (group3). Third, the number of sequence per group is really unbalanced.
It may comes from the fact that my time variable has a range of 123. I searched for articles that had an issue with a too long time range. I read in Sabherwal and Robey (1993) and in Shi and Prescott (2011) that you can standardize "each sequence by dividing the number of transformations required by the length of the longer sequence". How can I do that in R?
Please find underneath a description of my data:
library(TraMineRextras)
head(seq.tse.data)
seq.tse.data <- structure(list(
ID = c(1L, 2L, 2L, 3L, 3L, 3L, 3L, 3L, 4L, 4L, 4L, 4L,
4L, 4L, 4L, 5L, 5L, 5L, 6L, 6L, 6L, 6L, 6L, 6L,
6L, 6L, 7L, 7L, 7L, 7L, 7L, 7L, 7L, 7L),
Year = c(2008L, 2010L, 2012L, 2007L, 2009L, 2010L, 2012L,
2013L, 1996L, 1997L, 1999L, 2003L, 2006L, 2008L,
2012L, 2007L, 2007L, 2008L, 2003L, 2007L, 2007L,
2009L, 2009L, 2011L, 2014L, 2016L, 2006L, 2009L,
2011L, 2013L, 2013L, 2015L, 2015L, 2016L),
Event = c(5L, 4L, 5L, 3L, 1L, 5L, 5L, 5L, 3L,3L,3L,3L,3L,5L, 1L, 5L,
5L,5L,4L,5L, 5L, 5L, 5L, 5L, 5L,5L,5L,5L, 4L, 4L, 1L, 4L, 1L,5L)),
class = "data.frame", row.names = c(NA, -34L)
)
seq.sts <- TSE_to_STS(seq.tse.data,
id = 1, timestamp = 2, event = 3,
stm =NULL, tmin = 1935, tmax = 2018,
firstState = "None")
seq.SPS <- seqformat(seq.sts, 1:84, from = "STS", to = "SPS")
seq.obj <- seqdef(seq.SPS)
> head(seq.tse.data)
ID Year Event
1 1 2008 5
2 2 2010 4
3 2 2012 5
4 3 2007 3
5 3 2009 1
6 3 2010 5
> head(seq.obj)
Sequence
[1] (None,74)-(5,10)-1
[2] (None,76)-(4,2)-(5.4,6)-2
[3] (None,73)-(3,2)-(3.1,1)-(5.3.1,8)-3
[4] (None,62)-(3,12)-(5.3,4)-(5.3.1,6)-3
[5] (None,73)-(5,11)-1
[6] (None,69)-(4,4)-(5.4,11)-2
> head(alphabet(seq.obj),10)
[1] "(1,1)" "(1,10)" "(1,11)" "(1,12)" "(1,14)" "(1,19)" "(1,2)" "(1,21)" "(1,25)" "(1,3)"
...
[145] "(5.4.3.1,5)" "(5.4.3.1,6)" "(5.4.3.1,7)" "(5.4.3.1,8)" "(5.4.3.1.2,9)" "(None,1)" "(None,11)" "(None,20)"
[153] "(None,26)" "(None,30)" "(None,38)" "(None,41)" "(None,42)" "(None,44)" "(None,45)" "(None,49)"
[161] "(None,51)" "(None,53)" "(None,55)" "(None,57)" "(None,58)" "(None,59)" "(None,60)" "(None,61)"
[169] "(None,62)" "(None,64)" "(None,65)" "(None,66)" "(None,67)" "(None,68)" "(None,69)" "(None,7)"
[177] "(None,70)" "(None,71)" "(None,72)" "(None,73)" "(None,74)" "(None,75)" "(None,76)" "(None,77)"
[185] "(None,78)" "(None,79)"
Thanks in advance,
Antonin

Answer 1

I guess that your question is about normalizing the dissimilarities between sequences. E.g., Sabherwal and Robey (1993, p 557) refer to the distance standardization proposed by Abbott & Hyrcac (1990) and do not consider at all the standardization of a sequence. Anyway, I cannot figure out what the standardization of a sequence could be.
The seqdist function of TraMineR has a norm argument that can be used to normalize some of the distance measures proposed. Here is an excerpt from the seqdist help page:
Distances can optionally be normalized by means of the norm argument.
If set to "auto", Elzinga's normalization (similarity divided by
geometrical mean of the two sequence lengths) is applied to "LCS",
"LCP" and "RLCP" distances, while Abbott's normalization (distance
divided by length of the longer sequence) is used for "OM", "HAM" and
"DHD". Elzinga's method can be forced with "gmean" and Abbott's rule
with "maxlength". With "maxdist" the distance is normalized by its
maximal possible value. For more details, see Gabadinho et al. (2009,
2011). Finally, "YujianBo" is the normalization proposed by Yujian and
Bo (2007) that preserves the triangle inequality.
Let me warn you that while normalization makes distances between two short sequences (say of length 10) more comparable with distances between two long sequences (say of length 100), it does not solve the issue of comparing sequences of different lengths.
You find a detailed discussion on the normalization of distance and similarity in sequence analysis in Elzinga & Studer (2016).