On Mon, 11 Jan 2016 10:07:20 -0800, Marie Helweg-Larsen wrote:
Interesting article about new research in Economics showing how female
(but not male) economists are do not receive credit for their work if
they
publish with men (as opposed to publishing with women or by
themselves).
I wonder what a similar analysis in psychology would show. This pattern
did not show up in a similar analysis of sociology single or
co-authored
publications. The article suggests that one contributing factor might
be
that in economics authorship order is alphabetical (leaving the
relative
contribution uncertain). Check it out.
|
http://www.nytimes.com/2016/01/10/upshot/when-teamwork-doesnt-work-for-women.html?em_pos=small&emc=edit_up_20160111&nl=upshot&nlid=30609997&ref=headline&_r=0
I guess it goes without saying that one should check the original
sources that serve as the basis for a popular news article and the
main source for this article is an unpublished working paper by
a Harvard graduate student Heather Sarsons. Her paper can
be downloaded from her publications/research page at Harvard;
see:
http://scholar.harvard.edu/sarsons/publications/
Apparently, it was presented at this year's American Economic
Association (AEA) meeting on January 5, 2016, 10:15am session;
see:
https://www.aeaweb.org/aea/2016conference/program/preliminary.php?search_string=sarsons&search_type=last_name&association=&jel_class=&search=Search#search_box
The paper does not mention this.
Let me highlight a few points about the paper:
(1) It is a 40 page manuscript with12 pages of figures and tables
and one page of references, leaving 27 of text (including the
first page with the abstract). I mention this because Sarsons
does not describe her data until page 16 and there are some
issues that people should consider such as:
(A) The data come from only one Ph.D. granting institution and
this is identified as among the "top 30 PhD-granting universities
in the United States". How do we know it among the top 30?
The ranking come from the following table:
https://ideas.repec.org/top/top.usa.html
This suggests that the results only apply to elite institutions if the
university (presumably Harvard) is similar to other elite universities.
Whether it applies to non-elite institutions is an open question.
(B) The total sample was N=552 but the female sample was N=130
or about 24%. That is, this is a seriously "unbalanced" design which
can have a major effect on statistical analysis, especially if the
variances
are systematically different and/or the variances are related to the
sample size (this is generally known as the Fisher-Behren's problem;
for more info see the Wikipedia entry:
https://en.wikipedia.org/wiki/Behrens%E2%80%93Fisher_problem ).
Unfortunately, in Table 1 (where the first mention of the sample size
inequality is made, she says nothing about it in her text) she provides
only descriptive statistics but is not explicit which ones she is using
(presumably means and in parenthesis standard deviations) and
p-values from some unspecified statistical test -- it could all be
t-tests
with but this is unlikely for frequency data (2 x 2 table for gender by
tenure status, unless a test between proportions was done). It is
surprising that the statistical tests are not identified.
(C) An even odder aspect of the data in Table 1 is, if the numbers
in the parentheses are standard deviations, is that the standard
deviations are incredibly similar even though the male sample
is more than 3 times that of the female sample. This implies that
the amount of sampling error in the variance is not reduced as
the sample size increases (the variance of the Sampling Distribution
of Variances for N=422 has to be smaller than that based on N=130;
a variance = 3.61 in both samples will have a lower probability in
a sample N=422 than in a sample N=130).
(D) The problems with Table 1 are important to consider because
the first variable in the table is proportion of tenured faculty. The
overall/Full sample proportion is .71 with a standard deviation of 0.45
(if the number in parenthesis is the SD). In the male group, 0.77 or 77%
of the males are tenured (SD= 0.42) and in the female group 0.52 or
52% (SD=0.50) are tenured. The difference has a p-value of 0.001
for an unknown test and the point of the paper is to explain why there
is a difference of 0.25 or 25%. The two groups also differ on the
number of years to tenure (male M=6.6,SD=1.9, female M=7.1, SD=1.9),
and average rank of journal of published articles (male M= 46.3,SD=19.1,
female M=42.6,SD=17.3). A "marginally significant" result is that for
number of co-authored/multiple authored papers where p= 0.054
(male M=46.4, SD=19.1, female M=42.2, SD=20.1; females have
fewer co-authored/multiple authored papers -- think about that for
a bit). I don't think there was any correction for the number of tests
done.
(2) I won't go over the rest of the results (which include probit
regression)
because there really isn't enough information provided about what was
done to conclude that the analyses are valid, IMHO. Another point
to consider is that the sample spans the time period from 1975 to 2014
but there are no reported analyses involving year of data -- one assumes
that the number of female economics faculty has increased from 1975
to 2014 and that the overall conclusions may apply to only some of these
years but that's a whole 'nother can of worms.
It would be nice if the Sarsons makes the dataset publicly available as
well
as identifying which tests she used and which software package she used.
I can't tell if the lack of detail in the analysis is because (a) that's
what
economists do when they report results or (b) the author does not
realize that such information should be presented. In psychology, I
think such a manuscript would be considered seriously deficient.
As to whether a similar analysis can be done with psychology
data, it seems that it is possible but one has to decide whether one
wants to replicate this study (i.e., using one of the "Top 30
Universities"
that grant Ph.D.s in psychology -- should that be all degrees or a
subset like clinical psychology?) or are there more appropriate
designs and analyses to use instead? Anyone have a grad student
needing a dissertation topic? ;-)
As interesting as this research appears to be, especially as presented
in the NY Times article, it might be another example of premature
communication.
-Mike Palij
New York University
m...@nyu.edu
---
You are currently subscribed to tips as: arch...@mail-archive.com.
To unsubscribe click here:
http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=47832
or send a blank email to
leave-47832-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu
