Research:Knowledge Gaps Index/Measurement/Readers Survey 2023

This project aims to understand the demographics and motivations of Wikipedia Readers across language editions. It is part of the Knowledge Gaps Index focus on Readers of Wikipedias, and continues the work of the 2019 Readers survey.

Progress on this project can be followed at T341890.

Here is a summary of the 2023 Global Reader Survey results. For the full results please check the Results section.

• Wikipedia readers skew young, although this varies by project. However, readers aged 18-24 are a plurality of those 18+ in nearly all surveyed projects (the exceptions are dewiki and nlwiki).

• Wikipedia readers across all projects identify disproportionately as solely men. By project, readership of ukwiki, rowiki, and ruwiki are closest to gender parity.

• Wikipedia readers are highly-educated. Many readers are current students. Across each surveyed project, a majority of readers under age 30 are current students.

• Wikipedia readers are highly multilingual: a majority speak two or more languages fluently. However, readers of any given Wikipedia are overwhelmingly reading in a primary language. Readers of enwiki are most likely to be non-native language speakers.

• Consistent with previous research, we observe a pronounced gender gap in the length of reading sessions. However, an updated gender identity survey item allows us to specify that readers who identify solely as women read substantially fewer articles per reading session when compared to readers of any other gender identity.
• We observe some mixed evidence for gender-based differences in topical preferences. More research is needed to better understand this relationship.
• Watch this page for planned future analysis!

This project employed simple random sampling of Wikipedia readers using the QuickSurveys extension. The QuickSurveys opt-in was displayed to non-logged in users and asked them whether they would like to participate in a survey to help improve Wikipedia. Survey responses were collected using LimeSurvey, an external survey tool.

The goals of the survey are to make demographic estimates of Wikipedia readers across different language projects within the scope of the Knowledge Gaps Index, to understand motivations for reading Wikipedia, and to analyze whether there are differences by motivation and demographics in who reads which type of content.

Analyses of the survey data will primarily follow the 2019 edition of the survey.

Date	Milestone
September 22—October 4, 2023	enwiki pilot survey
November 14—November 22, 2023	enwiki full sample survey
November 30—December 14, 2023	arwiki, cswiki, dewiki, elwiki, eswiki, fawiki, frwiki, hewiki, hiwiki, jawiki, kowiki, nlwiki, plwiki, ptwiki, rowiki, ruwiki, trwiki, viwiki, ukwiki, zhwiki full sample surveys

This survey is governed by the Global Readers Survey privacy statement.

Surveys were fielded across 23 projects from November 14--December 18, 2023. A total of 80,242 complete survey responses were collected.

Global Readers Demographic Survey (2023) Fielding Summary

Project	Fielding Dates	QuickSurvey Sampling Ratio	Total LimeSurvey Initiations	Total Completes
arwiki (Arabic)	28/11- 18/12	12.4%	40526	5186
cswiki (Czech)	28/11- 11/12	10.0%	4592	1618
dewiki (German)	28/11- 11/12	5.2%	23797	9589
elwiki (Greek)	28/11- 13/12	20.0%	5557	1537
enwiki (English)	14-22/11	2.0%	40497	9479
eswiki (Spanish)	28/11- 18/12	6.6%	39071	8769
fawiki (Farsi)	28/11- 11/12	2.1%	7689	1850
frwiki (French)	28/11- 11/12	9.7%	23368	6617
hewiki (Hebrew)	28/11- 13/12	8.6%	5044	1609
hiwiki (Hindi)	28/11- 18/12	20.0%	62278	715
idwiki (Indonesian)	28/11- 13/12	15.0%	13671	1516
itwiki (Italian)	28/11- 11/12	2.5%	5855	1996
jawiki (Japanese	28/11- 18/12	1.6%	7023	1905
kowiki (Korean)	28/11- 18/12	20.0%	12800	1575
nlwiki (Dutch)	28/11- 18/12	7.5%	5420	1528
plwiki (Polish)	28/11- 11/12	7.5%	12341	3672
ptwiki (Portuguese)	28/11- 18/12	15.0%	32457	4619
rowiki (Romanian)	28/11- 11/12	21.1%	9820	2399
ruwiki (Russian)	28/11- 18/12	1.5%	15637	5357
trwiki (Turkish)	28/11- 11/12	7.5%	8568	1792
ukwiki (Ukrainian)	28/11- 11/12	6.4%	6576	2094
viwiki (Vietnamese)	28/11- 18/12	7.5%	6111	1075
zhwiki (Simplified and Traditional Chinese)	28/11- 18/12	7.1%	15841	3745

Only readers aged 18 years and older were considered eligible for the survey. As a result, all readers who opted into the survey were first shown an age-based screener question. Those who indicated they were under 18 had their survey sessions terminated.

Are you at least 18 years of age?
○ Yes
○ No

Unfortunately, legal protections for people under 18 mean we cannot survey you. Thank you for your interest!

Consistent with previous survey research conducted by the Wikimedia Foundation [1] [2] , we asked readers about their motivations for reading Wikipedia. However, in this survey, we allowed respondents to select multiple motivations and to write-in other motivations that were not listed as answer options.

I am reading this article because ...

Please select all answers that apply

□ I have a work or school-related assignment
□ I need to make a personal decision based on this topic (e.g., buy a book, choose a
travel destination)
□ I have a work or school-related assignment
□ I need to make a personal decision based on this topic (e.g., buy a book, choose a
travel destination)
□ I want to know more about a current event (e.g., a soccer game, a recent earthquake,
somebody’s death)
□ the topic was referenced in a piece of media (e.g., TV, radio, article, film, book)
□ the topic came up in a conversation
□ I am bored or randomly exploring Wikipedia for fun
□ this topic is important to me and I want to learn more about it (e.g., to learn about a culture)
□ Other:__________________________

When asked what motivated them to read the article they were sampled from during the 2023 Global Readers Survey, respondents were overall most likely to say the article topic was "personally important" to them. (Respondents were able to select multiple motivations).

Similarly, at the project level, readers of all surveyed projects except for Korean Wikipedia were most likely to say they were reading the article because it is personally important to them. Korean Wikipedia readers were most likely to say they were "bored or randomly exploring Wikipedia for fun".

Again, following previous survey research, we asked readers about the specific information needs that motivated them to read the article from which they were sampled.

I am reading this article to …
○ look up a specific fact or to get a quick answer
○ get an overview of the topic
○ get an in-depth understanding of the topic

Overall, Wikipedia readers are most likely to say they are are reading to "get an overview of the topic". However, reader information needs are fairly evenly distributed with 41.2% saying they are reading for an "overview", 32.1% to "look up a specific fact or to get a quick answer", and 26.0% to "get an in-depth understanding of the topic".

At the project level, Farsi Wikipedia readers are most likely to say they are looking for "an in-depth understanding" (52.5%), Hebrew Wikipedia readers are most likely to say they are seeking "an overview" (50.0%), and Vietnamese Wikipedia readers are most likely to say they need to "look up a specific fact or...get a quick answer" (42.2%).

We presented readers with the same survey question measuring their prior knowledge of the topic of the article they were reading that was used in previous readers surveys.

Prior to visiting this article … 
○ I was already familiar with the topic
○ I was not familiar with the topic, and I am learning about it for the first time

Overall, Wikipedia readers are more likely to say that they are already familiar with the topic they are reading about (55.0%) than not (44.2%).

Readers of most language projects are similarly more likely to be reading articles on topics with which they are already familiar. However, there are some exceptions: readers of Chinese Wikipedia are particularly likely to be reading on unfamiliar topics (59.5%). In contrast, Dutch Wikipedia readers are most likely to read on familiar topics (74.6%).

Age has been robustly associated with a broad range of social attitudes and behaviors^[1][2] (and even survey response quality^[3]) in addition to internet use and digital proficiency. Moreover, previous Wikimedia Foundation research has found that Wikipedia readers are disproportionately young. We measured age with the following item drawn from the Community Insights survey.

What is your age?

○ 18-24
○ 25-29
○ 30-39
○ 40-49
○ 50-59
○ 60-69
○ 70+
○ I prefer not to say

Of those 18 and older, respondents across all surveyed projects are most likely to be aged 18-24 (27.9% of readers 18+). However, the age distribution of readers varies considerably across the surveyed projects.

In particular, readers of Vietnamese Wikipedia are most likely to be under the age of 30 (61.5% aged 18-29), while Dutch Wikipedia (21.8% aged 18-29) and German Wikipedia (21.0% aged 18-29) readers are least likely to be under the age of 30.

Johnson et al. (2021)^[4] demonstrate key gender differences in Wikipedia readership; specifically, that men are overrepresented among Wikipedia readers and read more frequently and for longer sessions and that men and women show distinctive topical preferences. This is consistent with the well-known and persistent gender-based bias of Wikipedia content and persistent overrepresentation of men among Wikipedia editors.

In order to facilitate comparisons between surveys of Wikipedia readers and contributors to Wikimedia projects, this research employed a gender identity survey item aligned with that used in the 2024 Community Insights survey. Note that respondents to the arwiki, fawiki, and inwiki surveys were not presented with the "transgender", "non-binary", and "genderfluid" response options.

Which of these categories describe your gender
identity? Select all that apply.

□ Man
□ Woman
□ Transgender
□ Non-binary
□ Genderfluid
□ Other: _________________
○ I prefer not to say

Across all surveyed projects, a clear majority (63.3%) of respondents identified solely as men, 25.1% identified solely as women, 6.4% identified as genderdiverse, and 5.1% declined to provide an answer.

Readers identifying solely as men made up an outright majority in every surveyed project, but projects like Romanian Wikipedia (54.6% readers identifying as men only) and Ukrainian Wikipedia (51.7%) are substantially closer to gender parity than projects like Turkish Wikipedia (71.7%) or Indonesian Wikipedia (70.6%).

As summarized in the Taxonomy of Knowledge Gaps, a substantial body of research demonstrates that Wikipedia readers are disproportionately highly-educated. A related body of research suggests that English Wikipedia articles^[5] may not be readable for less-highly-literate readers, particularly for health-related content^[6][7], while more recent research suggests these findings can be extended to most other language versions.

Measuring educational attainment cross-nationally is a longstanding methodological challenge in survey research^[8]. This is further complicated in our case by the fact that Global Readers surveys are designed and sampled by language project rather than by geography (e.g., enwiki respondents alone are educated under a wide variety of very different educational systems). We also sought to balance survey item simplicity with cross-system comparability. Together, these constraints made it difficult for us to substantially localize our measures of educational attainment.

In this survey, we measured education with two survey items: one asking whether respondents were currently enrolled as students and a subsequent item asking non-students to indicate their level of educational attainment based loosely on the ISCED-1997 classifications. We employed this scheme rather than years of education completed as used in previous readers survey research to facilitate more direct comparisons both cross-nationally^[9] and with Community Insights data on contributors.

Are you currently enrolled as a student in school (for example, high school, vocational or trade school, a college or university)?
○ Yes
○ No
○ I'm not sure
○ I prefer not to say

Only shown to respondents who selected "No" above

What is the highest level of formal education you have completed?
○ I have no formal schooling
○ Some primary or elementary school
○ Primary or elementary school
○ Lower secondary or middle school
○ Upper secondary or high school
○ A post-secondary technical or vocational degree or certificate
○ A post-secondary or university degree
○ A post-graduate degree (e.g., master's, doctorate, or professional degree)
○ I'm not sure
○ I prefer not to say

Substantial shares of readers in every surveyed project indicated that they are currently enrolled students, although this varies considerably from fewer than one-in-five overall among Dutch (19.7%) and German (19.5%) Wikipedia readers to an outright majority of Vietnamese Wikipedia readers (54.2%).

In addition, current students comprise a majority of younger readers (those 18-29) in each surveyed project.

Overall, Wikipedia readers are highly-educated: a majority of non-students (56.0% total) have completed a Bachelors' degree (28.8%) or a post-graduate degree (27.2%).

At the project level, Indonesian Wikipedia readers are most likely to report an educational attainment at the upper secondary (high school) level or lower, while Polish Wikipedia readers are most likely to report holding a post-graduate degree.

Among non-students, Ukrainian Wikipedia readers (76.5%) are most likely overall to report having at least a Bachelor's degree, while Indonesian Wikipedia readers are the least likely (38.1%) relative to other surveyed projects.

In general, Wikipedia readers are highly multilingual. When asked what languages they speak fluently, fewer than half (44%) say they are fluent in only one language, while more than one-in-five (21.5%) say they speak three or more fluently. However, readers are overwhelmingly reading in (one of) their primary languange(s).

In all but one surveyed project, about nine-in-ten (or more) readers say they are reading in one of their primary languages. The relative exception to this finding is English Wikipedia, where more than one-in-four say English is not one of their primary languages.

In contrast, the prevalence of monolinguality in the project language varies considerably by project. In general, East Asian language projects (and Greek Wikipedia) show the highest levels of monolinguality among readers—especially readers of Japanese Wikipedia (90.1%). Conversely, readers of German Wikipedia (22.3%) and Turkish Wikipedia (22.8%) were least likely to say they were monolingual in the project language.

In order to measure cultural background gaps, as described in the Taxonomy of Knowledge Gaps we employ survey items adapted from the European Social Survey^[10] (and also used in the Community Insights survey of Wikimedia contributors) designed to measure whether respondents belong to:

• A minority ethnicity in the country where they live
• A group that is discriminated against in the country where they live
• Why they are discriminated against (if applicable)

Do you belong to a minority ethnic group in the country where you currently live?
○ Yes
○ No
○ I'm not sure
○ I prefer not to say

The UN Office of the High Commissioner for Human Rights (UNHCR), roughly estimates that 10-20 percent of the world population belongs ot a national, ethnic, religious, or linguistic minority. This is broadly consistent with our Global readers sample, where 15% of respondents indicate that they belong to an ethnic minority in the country where they live.

At the project level, readers of idwiki (20.7%) and enwiki (19.6%) are most likely to identify as an ethnic minority. Conversely, readers of itwiki (4.0%) and elwiki (3.7%) are least likely to identify as belonging to a minority ethnic group.

Sometimes people are discriminated against based on characteristics like abilities, physical appearance, or group belonging.

Would you describe yourself as a member of a group that has been discriminated against in the country where you currently live?

○ Yes
○ No
○ I'm not sure
○ I prefer not to say

One-in-four (25.0%) readers indicated that they belong to a group that is discriminated against in the country where they live. These findings are broadly similar to those reported in the 2023 Community Insights survey of Wikimedia contributors.

Readers of English wikipedia appear most likely to describe themselves as belonging to a discriminated group (31.9%). Readers of Vietnamese wikipedia are the least likely to identify that way (5.5%). Unfortunately, we are not able at this point to determine the extent to which project-level variation on this item is the product of different experiences, varying levels of willingness to identify as belonging to a marginalized group, or varying understandings of what it means to be discriminated against.

Readers who indicated that they belonged to a discriminated group were then asked to indicate on what grounds their identity/identities are discriminated against. Respondents were able to select as many as applied. Overall, readers were most likely to say they were discriminated against due to their gender (29.4%) or their skin color or race (27.6%).

This project employed simple random sampling of Wikipedia readers using the QuickSurveys extension. Sampling rates vary by project and are shown above. The QuickSurveys opt-in was displayed to non-logged-in readers only and asked whether they would consent to "Take a short survey and help us improve Wikipedia". We chose to employ the QuickSurvey tool to sample readers (rather than e.g., a Central Notice Banner) both for consistency with previous readers research conducted by the Wikimedia Foundation and to avoid sampling readers from non-article pages (e.g., talk pages, community pages, Wikipedia home pages).

Readers who consented to the survey were then linked out to a survey hosted on LimeSurvey, an open-source survey platform.

In order to account for sampling design and to better match the global population of Wikipedia readers, we apply weights based on global population parameters following the method described in DeBell and Krosnick (2009)^[11] implemented using the 'anesrake' software package written for R.

Survey responses were weighted at the project level by OS family (Android, iOS, Windows, other), referrer class (external via search engine, internal, other), session length (one, two, three or more), geography (weighting categories vary by project). For analyses at the global level, responses were also weighted by project shares of overall traffic during the time when the surveys were in the field.

Using webrequest data, survey responses can be linked to the pages each reader views during the reading session from which they were sampled. These pages can then be classified into one of 64 topics based on a language-agnostic topic classification method developed by Johnson, Gerlach, and Sáez-Trumper (2021)^[12]. Topic classifications recorded for each respondent are not mutually exclusive at either the article or reading session/respondent level. That is, any given article may be classified into multiple topics and respondents may read multiple articles per session.

At the most general level, Wikipedia articles are classified into one of four "top-level" topics: "Culture", "Geography", "STEM", and "History & Society". Overall, a majority of Global Readers Survey 2023 respondents (53%) viewed a Culture-related article. Geography-related articles were viewed by 43% of respondents, while readers were much less likely to view either STEM-related (19%) or History & Society-related (18%) articles.

Next, the figure below depicts the proportion of readers who read articles on a given topic during the reading session from which they were sampled for the survey. Only "bottom-level" topics are shown to facilitate like-to-like comparisons. That is, because all Geography articles that are about the region within the Americas known as North America are also simultaneously Geography articles, Geography articles about a region, and Geography articles about a region within the Americas, we only show statistics for "Geography.Regions.Americas.North_America".

Overall, the most frequently-read Geography topic is "Geography.Regions.Americas.North_America" (visited by 10.3% of readers). The most-frequently read Culture topic is "Culture.Media.Music" (visited by 7.7% of readers), while the most frequently read History and Society Topic is "History_and_Society.Politics_and_government" (visited by 5.0% of readers) and the top STEM topic is "STEM.Biology" (visited by 3.8% of readers).

Building on findings from Johnson et al. (2021)^[4], we examine how the way readers interact with Wikipedia varies by their gender identities.

For this research, we define session length as the number of webrequests logged within a reading session--including pages viewed both before and after the survey response. This follows the method described in Singer et al. (2017) and utilized in the 2019 Global Readers survey.^[13] In other words, session length is measured in terms of articles viewed, not in time elapsed. Overall, the weighted mean session length for 2023 Global readers survey respondents across all projects was 4.2 articles viewed.

Consistent with findings reported by Johnson et al. (2021)^[4], we again observe a significant and substantial gender gap in session length, where readers identifying as women view fewer articles per reading session. However, the 2023 Global Readers Survey introduced a new gender identity item, allowing respondents to select multiple gender identities. Thus, findings from this research suggest that the gender gap in session length may be due to readers who identify solely as women reading distinctively fewer articles per reading session.

Similarly, when genderdiverse identities are shown as non-mutually-exclusive categories, it becomes clear that it is readers who identify solely as women who have distinctively shorter reading session lengths. This should complicate our understanding of the nature of the Wikipedia readership gender gap—and how to close it. For example, product and policy interventions aimed at closing the gender gap may have to specifically target readers who identify solely as women.

Johnson et al. (2021)^[4] find that readers identifying as women are generally more likely to read biographies of women compared to readers identifying as men (although men, as a majority of Wikipedia readership nonetheless comprise an absolute majority of readers of women's biographies). We find a similar pattern in the 2023 Global Readers Survey: readers identifying (solely) as women are about 1.65× as likely to read women's biography articles as readers identifying (solely) as men. The figure below shows the probability that a reader will read a woman's biography article conditional on their gender identity. For example, 2.7% (±1.4%) of readers identifying as genderdiverse viewed a biographical article about a woman during the reading session from which they were sampled for the survey.

However, some of the apparent "self-focused" gender gap in readership for women's biography articles is due to a broader gender gap in interest in biographical articles in general. When we focus only on those readers who viewed any biographical article in the course of their reading session, the gender gap begins to close: readers identifying (solely) as women are about 1.3× as likely to read women's biographies as readers identifying (solely) as men.

What's more, when we incorporate covariate adjustments for session length, age, access method, educational attainment, and urbanity in a logistic regression predicting P(Reading women's biography), gender identity is not a statistically significant predictor. However, interpreting this finding is complicated by the fact that e.g., age, educational attainment, and urbanity are also likely to be (although this is unknowable with our data) systematically related to survey response propensity, raising the possibility that adjusting for these covariates amounts to conditioning on a collider. Thus, further research maybe required to better identify (and quantify) the extent of potential systematic nonresponse bias (e.g., by comparing demographic data collected via different sampling methods such as QuickSurvey, QuickSurvey link-out to Limesurvey, and Central Notice Banner link-out to Limesurvey).

We plan to conduct the following further analyses of the 2023 Global Readers survey data and to share their results here:

• Additional work to better understand the relationship of gender identity to topical preferences building on Johnson et al. (2021)^[4].
• Analysis linking educational attainment with language-agnostic article readability scores developed by Trokhymovych et al. (2024) ^[14]
• A comparative analysis of reader demographics from this study with data collected through a central notice banner-sampled study of German, English, Spanish, French, Italian, Portuguese, Turkish, and Ukrainian Wikipedias conducted by Cruciani et al. (2023)^[15] in May-June of 2023.