Wills et al. (2009), Exp. 1A - clarifications
=============================================

During the compilation of this DAU in 2016, I discovered a
number of places in Wills et al. (2009, JCP) where the description of
the Methods and Results of Experiment 1A was poor. This document fills
in the gaps and corrects the errors.

Method
------

1. The observing response is on a FT 10s schedule.

2. The schedule on the response keys is "tand FT 2s VI 0.5s".  VI 0.5s
   is implemented as a random selection, on each trial, of 0 seconds
   and 1 second.

3. Phase 1 training started with a session length of 80 trials. Birds
   were subsequently moved to a 64-trial session length for phase 1
   later on.

4. In some Phase 1 sessions, a forced-choice procedure was used on
   most or all trials. This is where the pigeon cannot proceed to the
   next trial until it pecks the correct key. Forced-choice trials are
   indicated in the raw data file in column 'fc' by the presence of an 'F'.

5. References in Wills et al. (2009) to number of sessions completed
   in phase 1 do not count 80-trial sessions, forced-choice sessions,
   or some other sessions. For archival purposes, the session
   numbering as used in the original paper is in column 'lsess'. The
   column 'sess' provides  a simple incremental count of sessions
   started. Column 'sess' is used in the current analyses.

5. In rare cases, there was up to three weeks between one session and
   the next. 

Technical / experimenter errors
-------------------------------

6. There was a technical error that led to two birds being passed to
   phase 2 without meeting the stated criterion (details in Results
   below). The technical error can be reproduced by using column
   'snum' in the data file. 

7. Three birds (At, Io, Ly) had an additional test session. These are
   included the raw data file.

8. In some sessions, one or both of the food delivery hoppers
   malfunctioned. These sessions are included in the raw data file.

9. For one session (As, 2006-07-03), the data did not record. This
   session is indicated in the raw data file.

Results
-------

Phase 1 - Although every bird had an accuracy of at least 75% on their
final phase 1 session, two birds (At, Ly) did not pass the criterion
of being at least 75% accurate on both levels of at least three
dimensions (they both passed it on two dimensions). This was due to a
technical error. One can reproduce this error by using column 'snum'
in your analysis. For all other purposes, use column 'stim'. 

Phase 2 - The criterion is 87.5%. All birds reach the criterion,
taking between 1 and 9 sessions to do so.

Test sessions - The excess test sessions (see Experimenter errors,
above) were removed prior to analysis (session numbers - At: 76, Io:
63, Ly: 67). Table 1 of Wills et al. (2009) contains some arithmetic
errors, a corrected version is shown below. In all but one case, the
identified strategy is unchanged. For bird Pn, the reported Overall
Similarity strategy ties with a Unidimensional (Donut) strategy. This
change does not affect the conclusions reported in the paper. In
particular, the birds in Experiment 1A and the humans in Experiment 1B
do not differ significantly in the proportion of OS sorters
observed. The trend remains for OS to be more prevalent in humans than
in pigeons, contrary to predictions of Differentiation Theory.

Table 1 (replacement): Percentage of one-away test trials.

 subj bar chk loz don os    win
   As  39  61      58 58    chk
   At  39  72      50 61    chk
   Ba  54  51      46 51    bar
   Io  65  46      58 71     os
   Le      44  47  72 64    don
   Ly  53  64      50 67     os
   Pn  49  51      62 62 os/don
   Sf  62  49      54 65     os


Andy Wills
2016-02-22