Talk:Mind the Gap

Problems

I know of at least 2 people who were provided the survey and are male in real life. I think if you want to have a survey that adequately about women, you can't have any men involved even if they pretend to be female online. I'm sure that many women would think that it would be a problem if the results were skewed in anyway because of people who are not actually women. It is one of the problems with allowing people to create their own identities and then trying to get statistics off of them. Ottava Rima (talk) 15:28, 7 November 2011 (UTC)

It's possible for men to lie about their gender, just as it's possible for women to lie about their gender as they might want to do if they feel at risk of harassment. We cannot prevent people from lying, so we do the best we can and hope that the lies are rare enough and/or mutual enough that they are not significant for the outcome of the analysis. Pine^talk 06:54, 21 November 2011 (UTC)

Future directions

Hi! This was great reading, and fits in nicely with thoughts I've had about why people will engage with Wikipedia editing. It works extremely well as a discussion paper, which I assume was the aim. In reading it, though, I had a few thoughts on how it could be further developed, so I'm tossing them here on the grounds that it seems like the right place. :)

My main concern for the first part is the reliance on the Women and Wikimedia Survey 2011. I think Ottava has a valid point, but more fundamentally, the 2001 survey primarily looked at those who self-identify as women. Thus the figure of 9% quoted is not in relation to all female editors, but only in relation to self-identified female editors. I don't know if this is the case, but it might be possible that more people willing to self identify belong to a particular group, and thus self identified female editors on Wikipedia are not a representative portion of female editors on the whole. (There's also an error in "lesbians account for 24% of the contributors", which contradicts the rest of the statistics). The other concern with the 2011 survey is that it also allowed participants to invite others to answer the survey. This will tend to skew the results, as people will tend to ask others whom they know, possibly through shared interests. Thus you might get an over representation of one or more groups because of a tendency for people within those groups to invite others who are similar, rather than those from outside.

Because of that, I'd try and drop the reliance on the 2011 survey - it serves to point to what sorts of questions a full survey could do, and it does that really well, but the methodology needs to be refined before the data collected can be used as more than a discussion point.

I'm also a tad nervous about the "57% of Wikipedia's female contributors are single", as it seemed like an interesting figure, but according to ABS stats, 49% of adults in Australia are not in a relationship. In conjunction with the age range of the survey, which included people from the age of 12 and with a medium age of 31, that figure seems unsurprising taken on its own.

More generally, and perhaps more fundamentally - the tool used here seems to be based on the assumption that authors have the opportunity to write in their natural style. The study was based on single-authored texts, not, as is the case of Wikipedia, collaborativly authored texts. So I'm curious about the extent to which writing style is influenced by the purpose of the writing (writing for an encyclopedia would have specific style requirements), and by other external factors such as the MoS. I'd love to see more about how that may influence the figures. Unfortunately, without knowing the extent to which that influences the results, the comparison with other Wikis doesn't provide a lot of data. But it does point to some interesting future research.

Anyway, just some thoughts. As mentioned, it is in keeping with what I would expect to find, and I presume that this was intended to spark discussion, hence my response. :) There's certainly a lot worth discussing in the research. - Bilby 06:28, 8 November 2011 (UTC)

That's a good point about the proportion of "singletons" in the general population -- it is growing in Western countries. Maybe it would be more telling to ask whether or not female editors are raising children. I think that would give a better indication of how much available free time they have to do anything online, and how they prioritize their time. Food for thought. OttawaAC 23:46, 14 November 2011 (UTC)

Issues with identifying people's gender based on text style

Am I the only one who has reservations about this idea? I understand that large-scale studies have identified some words that are more associated with one gender, but how reliable are these 'author gender tests' really? I'd like to see some evidence that they are actually generally correct before accepting the results of this study at face value. The fact that they apparently identified all of the articles on the GeekFeminism wiki (a wiki with a large, if not overwhelming, proportion of female contributors) as written by men raises doubts about their reliability. I'm inclined to think that these tools are not measuring gender at all, but rather style/tone of writing, and the words they identify as 'male' are associated with a more encyclopaedic tone. So it's no wonder Wikipedia scores highly on such measures. Robofish 23:58, 16 November 2011 (UTC)

The very idea smells sexist, indeed. Measuring style/tone can still be meaningful and useless, though, as the page notes. Nemo 09:14, 17 November 2011 (UTC)

What struck me is the revelation that other wikis with much greater female participation rates -- the Geek Feminism wiki in particular -- actually have comparable writing styles (under this metric) to Wikipedia. I'd like to see a larger sample size from each Wiki, of course, but with the results given here, it tells me that the writing style gender assignments are either fatally flawed, or inherent somehow in the nature of information writing. LtPowers 15:36, 20 November 2011 (UTC)

The importance of other wikis, for me, was these other wikis are just like Wikipedia: They require the main text be written using a male writing style BUT these wikis still attract a large female editor base. Ultimately, for me, when coupled with the user page information, they confirm several things that Wikipedia's female population: 1) Wikipedia's female editor is not representative of female writers in general, 2) Wikipedia's existing female writer base may actually be harmful to the goal of reducing the gendergap because of how they participate and literature that suggests some of these women may drive off other female participation, 3) "The nature of Wikipedia using male information styles" cannot be an excuse because other Wikis can write in male styles and attract large amounts of female participation. --LauraHale 19:27, 20 November 2011 (UTC)

Possible re-interpretation of conclusions

This article concludes that Wikipedia attracts women who write like men (also, can we perhaps use the term women and men, rather than males and females? the latter two are used to oppress trans-gendered men and women and that is just ht beginning of the problem). My interpretation of the conclusions would be more that the edits which are allowed to stand, and the women who are accepted as part of the Wikipedia community are more likely to read as masculine. I suspect feminine writing is more likely to be challenged on NPOV, and that the sexism in online communities is likely to be more acutely felt by women who write using a feminine style, and as such I suspect Wikipedia is the cause of the masculine writing style, rather than its effect.

We chose the terms male and female, instead of men and women, because there are people under the age of 18 who edit Wikipedia. We felt that the terms male and female were more accurate. Pine^talk 06:58, 21 November 2011 (UTC)

I also see many ways to interpret the conclusions and would like much more detailed discussion of them. There is a logical circularity to saying that you can, on the one hand, distinguish male and female writing styles, but on the other hand, women on Wikipedia write like men. Starting from the resulting data, one might conclude more simply that there are not pronounced gender differences in Wikipedia writing, even if there are differences in formal BNC writing, which is certainly neither a logical nor empirical absurdity. (That does not mean I disagree with your interesting inference in a different direction; you may well be correct, but I think substantially more support is needed.) The fact that the sample from which you draw your base observations about gender difference (formal writing from the British National Corpus) is very different from the user pages and/or a subset of gender-related Wikipedia pages is also troubling. The fact that Wikipedia pages are collaboratively edited, as the user above suggests, whereas BNC articles typically have much more traditional editing styles, also suggests that you may be comparing apples and oranges. I don't see a strong effort to distinguish between edits and original text contributions, for example.

This is an important contribution to an ongoing discussion that is widely covered in the academic literature; I would like to see a much closer integration of your methodology with current techniques of corpus analysis, including submission to a peer-reviewed journal, where you will receive more feedback from corpus linguists who can guide you to additional resources, methods, authors, and experiments that bear on your question. As presented now, there are too many unchecked assumptions and inferences without methodological backup to draw any strong conclusions from your obviously interesting data and work. I appreciate Wikipedia's DIY ethic, but I am not clear how the persistent side-stepping of existing research strands, and reliance on just a few (if important ones), helps your audience to have confidence in your results. Have you, for example, run the paper by Shlomo Argamon, Moshe Koppel, Jonathan Fine, and Anat Rachel Shimoni, whose methods and research you are relying on heavily? I think they would have a lot of advice and support for your efforts. [[[Special:Contributions/173.53.108.220|173.53.108.220]] 17:14, 20 January 2012 (UTC)]

Wording differences on Wikipedia between male and female user pages

This isn't a fully processed thought, but rather a data sample based on what we've already done (and that wasn't included in the "paper" largely because it was getting unwieldy and there didn't appear to be much added value to putting this into the article beyond YUM! TASTY DATA! YUM!). Using the original list of self identified male and female contributors from user boxes as in the original paper, there were 867 females represented in the female data set, and 568 males represented in the male data set. For each data set, we ran a program that counted the total number of occurrences of all words for all females and all males. Example: Female 1 uses the word pornography 5 times, baseball 2 times and love 15 times. Female 2 uses the word pornography 0 times, baseball 20 times and love 6 times. Total for all female scores: pornography 5, baseball 22, love 21 times. All words with 3 or fewer letters were removed from the list. All words with _ in them were removed. A number of words that implied html or wiki specific coding were removed. These included words like table, alignleft, alignright. (Not all were removed, and some may have remained. Important to remember, all these words that were not included were not included for both genders.)

Women had 10,871 unique words they used. Men had 8,933 unique words they used. The table below is the top 100 words used by according to each gender.

Female word list	Female Use count	Female word rank	Male word list	Male use count	Male word rank
this	864	1	this	947	1
that	621	2	wikipedia	414	2
wikipedia	595	3	that	381	3
have	559	4	image	335	4
with	412	5	white	319	5
page	393	6	have	252	6
about	378	7	page	251	7
from	295	8	with	245	8
like	293	9	info	232	9
image	285	10	wiki	224	10
userboxes	285	11	from	223	11
also	273	12	articles	220	12
articles	273	13	user	218	13
white	270	14	nbsp	207	14
wiki	249	15	userboxes	177	15
user	223	16	about	171	16
more	210	17	article	145	17
info	208	18	middle	138	18
here	207	19	small	137	19
navframe	201	20	also	133	20
just	200	21	time	133	21
will	197	22	here	130	22
time	195	23	your	128	23
other	188	24	transparent	124	24
some	188	25	like	120	25
love	181	26	green	115	26
small	161	27	navframe	109	27
when	157	28	more	103	28
article	155	29	other	101	29
been	150	30	please	101	29
family	149	31	some	100	31
school	148	32	created	98	32
people	142	33	work	98	32
know	141	34	family	96	34
there	138	35	list	95	35
music	134	36	university	94	36
things	133	37	will	93	37
nbsp	127	38	interests	92	38
english	125	39	history	85	39
much	125	40	talk	84	40
middle	123	41	there	83	41
your	123	42	gray	82	42
interests	122	43	music	82	43
display	121	44	people	80	44
what	120	45	contributions	78	45
currently	117	46	been	77	46
pages	115	47	english	77	46
because	114	48	information	77	46
cellpadding	112	49	just	77	46
please	111	50	edits	76	50
auto	107	51	barnstar	74	51
editing	107	51	welcome	74	51
well	107	51	display	73	53
hello	106	54	supports	73	54
most	106	54	currently	71	55
navhead	105	56	ffffff	71	55
really	105	56	know	69	57
talk	105	56	cellpadding	68	58
navcontent	104	59	weight	68	58
read	103	60	hello	67	60
favorite	102	61	what	67	60
university	102	61	only	66	62
information	101	63	which	66	62
which	101	63	when	65	64
work	101	63	free	64	65
good	100	66	most	64	65
normal	99	67	united	64	65
since	98	68	would	64	65
they	97	69	category	63	69
very	97	69	pages	63	69
would	97	69	position	63	69
only	96	72	plainlinks	61	72
than	96	72	they	61	72
contributions	95	74	eats	60	74
help	95	74	school	60	74
history	94	76	world	60	74
life	94	76	auto	59	77
list	93	78	first	59	77
make	91	79	editing	58	78
many	91	79	make	58	78
years	91	79	since	58	78
serif	88	82	computer	57	82
barnstar	87	83	good	57	82
should	87	83	blue	56	82
wikimedia	86	85	live	56	82
feel	84	86	navcontent	56	82
ffffff	84	86	navhead	56	82
something	84	86	data	55	88
born	82	89	inherit	55	88
face	82	89	where	55	88
find	82	89	life	54	91
sans	82	89	orange	54	91
anything	81	93	party	54	91
first	81	93	related	54	91
live	80	95	serif	54	91
writing	80	96	things	54	91
reading	79	97	large	53	97
student	78	98	years	53	97
interested	77	99	collapsible	52	99
want	77	99	games	52	100
welcome	77	99	than	52	100
them	75	102	them	52	100
think	75	102	born	50	103

Is there a lot of difference between ranking of words on these lists which suggest different patterns in gender? 77 words appear on both lists for the top 100. They are: that, wikipedia, have, with, page, about, from, like, image, userboxes, also, articles, white, wiki, user, more, info, here, navframe, just, will, time, other, some, small, when, article, been, family, school, people, know, there, music, things, nbsp, english, middle, your, interests, display, what, currently, pages, cellpadding, please, auto, editing, hello, most, navhead, talk, navcontent, university, information, which, work, good, since, they, would, only, than, contributions, history, life, list, make, years, serif, barnstar, ffffff, first, live, welcome, them.

Words that appeared on the female list of top 100 words but not the male list include: love, much, because, well, really, read, favorite, normal, very, help, many, should, wikimedia, feel, something, born, face, find, sans, anything, writing, reading, student, interested, want, think. These words may appear on both lists anyway. For example, born is ranked the 90th most used word for women and the 103rd most used word for men. Help ranks 118 for men, and 74 for women. (Helpful is the 1192 most used word by women, and 1288 for men. Helping ranks 603 for men and 707 for women.) Without doing a lot of math, this data appears to confirm the conclusion for other data that male and female word usage on Wikipedia does not differ substantial. It appears highly unlikely that a word list could be used to identify the gender of Wikipedia users based on the text they write against a body of work based on known gendered Wikipedians. --LauraHale 04:42, 21 November 2011 (UTC)

And because the data was sitting around… a comparison of the top one hundred words used based on the users and there stated gender/program identified gender.

Stated F identified F Count	Stated F identified M Word	Stated F identified M Count	Stated M identified F Word	Stated M identified F Count	Stated M identified M Word	Stated M identified M Count
282	this	705	wikipedia	113	this	894
253	that	338	that	95	image	329
235	wikipedia	334	have	80	wikipedia	306
219	have	304	wiki	77	white	297
169	image	272	with	65	that	293
164	page	239	like	61	info	218
162	white	230	page	59	nbsp	200
136	about	201	articles	58	page	193
124	with	194	from	58	with	181
107	from	190	user	54	have	172
106	also	176	this	53	articles	166
106	articles	167	about	52	from	164
101	userboxes	161	data	48	user	164
100	info	153	label	47	wiki	146
100	user	149	userboxes	47	userboxes	131
99	wiki	145	also	40	middle	123
96	navframe	135	transparent	37	about	121
96	like	130	here	33	article	119
96	small	127	more	33	small	108
90	some	127	please	33	your	108
83	other	122	collapsible	31	green	106
77	more	111	auto	30	time	101
74	middle	108	edits	30	here	98
71	here	106	just	30	navframe	95
71	nbsp	103	talk	30	also	92
69	family	101	information	29	transparent	85
69	time	98	inherit	29	work	82
65	will	96	list	29	other	81
63	article	94	small	29	some	79
63	just	94	time	29	created	75
63	school	89	university	25	family	75
61	display	88	will	25	gray	74
60	english	85	english	24	more	70
58	love	85	music	24	supports	70
56	currently	84	welcome	24	university	69
55	people	83	dcdcdc	23	will	69
55	barnstar	82	interests	23	barnstar	68
52	music	82	school	23	please	68
52	been	81	white	23	interests	67
51	pages	79	because	22	list	67
49	interests	78	currently	22	history	66
49	read	75	know	22	there	66
49	serif	73	very	22	display	62
48	what	73	been	21	ffffff	61
48	face	72	category	21	eats	60
48	navcontent	72	family	21	people	59
47	navhead	72	hello	21	cellpadding	58
47	your	71	people	21	like	58
46	cellpadding	70	some	21	music	58
46	auto	69	years	21	what	58
46	favorite	69	your	21	been	56
46	there	69	article	20	contributions	56
46	sans	68	contributions	20	position	56
45	well	68	created	20	talk	55
45	university	67	history	20	weight	55
44	normal	66	only	20	english	53
43	work	66	other	20	blue	52
42	georgia	65	languages	19	united	52
41	information	65	love	19	free	51
41	list	65	make	19	information	50
40	gray	64	much	19	orange	50
40	history	63	there	19	welcome	50
40	hello	62	they	19	which	50
40	when	62	email	18	currently	49
39	only	61	help	18	navcontent	49
38	things	61	infobox	18	navhead	49
38	should	60	most	18	pages	49
38	contributions	59	myself	18	world	49
38	good	59	nowiki	18	hello	48
37	most	59	work	18	large	48
36	than	58	made	17	mijzelffan	48
36	which	58	student	17	when	48
36	wikimedia	58	when	17	would	48
35	editing	57	leave	16	first	47
35	logo	57	message	16	just	47
35	born	56	pages	16	know	47
34	film	56	related	16	life	47
34	first	56	which	16	airborne	46
34	ffffff	55	would	16	computer	46
34	please	55	education	15	most	46
34	know	53	high	15	only	46
33	site	53	into	15	plainlinks	46
33	years	53	male	15	edits	45
33	both	52	many	15	flag	45
33	much	52	plainlinks	15	than	45
32	cornflowerblue	51	politics	15	where	45
32	since	51	religion	15	editing	44
32	student	51	since	15	gold	44
32	writing	51	things	15	good	44
32	high	50	editing	14	live	44
32	life	49	everything	14	serif	44
31	they	49	info	14	born	43
31	topright	49	navframe	14	party	43
31	american	48	them	14	since	43
30	female	48	vandalism	14	wrote	43
30	find	48	bgcolor	13	best	42
30	interested	48	editor	13	category	42
30	something	47	interested	13	games	42
29	verdana	47	movies	13	they	42
29	supports	46	weight	13	make	41
29	welcome	46	world	13	absolute	40
		would	46			space	40
						well	40

Here, there appear to be a few differences. --LauraHale 07:17, 21 November 2011 (UTC)

Female wikiHow users as a benchmark for female Wikipedians: Female Wikipedians use more male styles of writing than female wikiHowians

I used the user boxes on wikiHow to create a list of wikiHow users by gender. I then ran their userpages through the same analysis I did for Wikipedia users where we knew their gender based on user boxes. Basically, the same methodology described on the main page, only switching sites. There were 256 unique users who had gender boxes identifying them by gender. When scores were removed for people scoring the same such as 0-0 (and effectively being gender neutral or undetermined), there were 231 users left, with 25 having been removed. The results are here:

Category	Gender	Count	Percentage
Wikihow Identified Gender	Male	66	28.6%
.	Female	165	71.4%
Program identified gender	Male	131	56.7%
.	Female	100	43.3%
Correctness	Yes - Female	81	49.1%
.	Yes - Male	47	71.2%
.	No - Female	84	50.9%
.	No - Male	19	28.8%

We're dealing with a much smaller sample from a site that has 43% female participation but an overall smaller contributor base. They have a greater percentage of females identifying with userboxes than Wikipedia does. (Suggests to me that women feel even more comfortable expressing femaleness there.) Beyond that, the program correctly identifies women 11% more often than it does on Wikipedia. 11% in this case seems significant to me. I think we can assume some bias for male writing nature in both spaces because of the factual nature conveyance for both… but Wikipedia's population and WikiHow's should match up if they were attracting similar types of females… and they aren't because there is that 11% difference. Beyond that, the program correctly identified male users within 1% (both roughly 71%) for wikiHow and Wikipedia. This says to me the two wikis have unique groups of female users as their characteristics are not the same. Wikipedia's females use male language much more in their personal space than wikiHow's female contributors.

Also worth noting, the female users appear to have their male/female word count points grouped more closely together than their male counterparts. Difference in STDEV for females between male and female scores is 77. For males, the difference is 188. Difference for women for mean between the male score and female score is 44. For men, it is 135. I'm guessing that we have a situation, where when plotted, the men would be less close to the gender neutral line than the women would be. Better idea of the users with in the group…

STDEV	Female score	Male score
Female user	1380.1	1457.5
Male user	824.5	1012.9
MEAN	Female score	Male score
Female user	810.1	852.3
Male user	558.5	693.6
MEDIAN	Female score	Male score
Female user	324	367
Male user	276.5	319.5
MODE	Female score	Male score
Female user	0	0
Male user	0	213

All this new data supports the methodology being valid. You may not like the study about gendered language and the word lists, but they developed a method of determining different patterns of language usage between genders. The wikiHow data really ,really supports the validity of it. Wikipedian female users over representing as male supports the supposition that female Wikipedians are much more likely to use male coded language and they are not representative of the wider of females. There is no real valid place to critique the methodology as flawed, because the distinct groupings validate it, and the wikiHow data is just the icing on the cake. --LauraHale 11:10, 21 November 2011 (UTC)