[Babase] One user's suggestion on rank data
Jeanne Altmann
altj at princeton.edu
Tue Oct 10 14:29:16 EDT 2006
Hi Russ,
Thanks for the suggestion based on your perspective as a user of the
ranks tables. Ranks are our most derived data in BABASE. What one
would consider the raw data are the matrices for each month, which are
also in BABASE, and those are what someone might like to take the time
to work with in situations as you describe, not a trivial project! No
single number would be useable as the sample size in determing any
individual's rank, to a considerable extent because it depends on sample
sizes of all other individuals as well, so one would probably want to
develop a set of criteria and metrices to use to create some sort of
uncertainty measure, probably one that would depend on the project at
hand.
-----Original Message-----
From: babase-bounces at eeblistserv.princeton.edu
[mailto:babase-bounces at eeblistserv.princeton.edu] On Behalf Of Russ Van
Horn
Sent: Tuesday, October 10, 2006 1:07 PM
To: The Baboon Database Project
Subject: [Babase] One user's suggestion on rank data
Hi all,
This may be a bit off the ranker topic, but I have a related
suggestion that I think is worth considering during the discussion of
the ranking system. This comes from the perspective of someone who uses
rank data from time to time, but who doesn't fully understand the
detailed process by which ranks are generated from the observations of
agonistic interactions.
After ranks are assigned, as the ranks are added to BABASE, I think
it would be useful for those conducting analyses with ranks to be able
to assess how much data went into the assignment of those ranks.
At present it's not straightforward for someone like myself to make such
an assessment. Would it be possible to add a new field to what is now
the 'ranks' table in FoxPro, presenting the number of agonistic
interactions used to generate each rank?
Here's why I think it would be worth adding another field, possibly
to 'ranks'. To me, as a user of the system, assigned ranks are estimated
data. Regardless of the type of assigned rank (e.g., ADM, ALL), and
regardless of whether I'm using assigned rank as a predictor variable or
as a response variable, there might be data points (i.e., ranks) which I
might want to treat as more reliable than others. (I'm proceeding from
the assumption that rank can be more accurately and precisely estimated
if more agonistic encounters were observed.) For example, suppose I
conduct an analysis of whether adult male rank influenced the
probability of secondary dispersal, or the probability of predation on
gazelle fawns, or the probability of paternity, etc. Whatever the
response variable, I might find that rank was a good predictor during
most time periods or for most males, but that rank wasn't a good
predictor for some time periods or males.
If all adult male ranks were based on comparable numbers of observed
agonistic encounters, then I might plausibly try to identify factors
other than rank that might be influencing secondary dispersal, or
predation, or paternity, etc. However, if the predictive power of rank
was low only during time periods, or for males, for which there weren't
many observed agonistic encounters, then the possibility exists that I'm
seeing a spurious result, and that the general predictive model is
valid. (And, that I should kick myself for not including sample size as
a covariate.) At present it's difficult for users not involved with rank
generation to determine whether, in fact, all ranks are created equal.
Cheers,
Russ
--
........................................................................
..............................
R. C. Van Horn, Ph.D.
Research Associate
Department of Biology
Box 90338 phone: 919-660-7297, 660-7306
Duke University email: rvanhorn at duke.edu
Durham, NC 27708-0338 web: http://www.duke.edu/~rvanhorn
........................................................................
................................
_______________________________________________
Babase mailing list
Babase at www.eco.princeton.edu
http://www.eco.princeton.edu/mailman/listinfo/babase
More information about the Babase
mailing list