Poster presentation at Southeast Data Librarian Symposium 2022, Thursday,October 13 Link
This poster proposal seeks to address data sharing, positionality, and diversity in research.
Researchers create data and share that data in university repositories. Once data is accessed, downloaded, and used in other research
it loses context. Data is not neutral. To understand data, it's important to understand the standpoint from which it was collected.
Standpoint Theory tells us that research perspectives are shaped by social position, and that knowledge (and data) is not objective but contextual.
Positionality statements aim to place the research in the context of the researcher's social position.
A data positionality statement written from the standpoint of the data would give context to the data itself — the who, what, where, and why.
We can use the positionality statements to make this information more accessible (metadata). This context matters. We all want more diversity in research
materials, but how do we get it and how do we find it? We have to label it. This work has implications for the reuse of data, machine learning/data science,
and a bigger picture look at the context of all this data.
"We saw in Chapter 1 that the feminist empiricist strategy argues that sexism and androcentrism are social biases, prejudices based on false beliefs (caused by superstitions, customs, ignorance, and miseducation) and on hostile attitudes. These prejudices enter research particularly as the stage of identification and definition of scientific problems, but also in the design of research and the collection and interpretation of evidence. According to this strategy, such biases can be eliminated by stricter adherence to the existing norms of scientific inquiry. Moreover, movements for social liberation make it possible for people to see the world in an enlarged perspective because they remove the covers and blinders that obscure knowledge and observation." (Harding, 1986, p. 151)
"A social history of standpoint theory would focus on what happens when marginalized peoples begin to gain public voice. In societies where scientific rationality and objectivity are claimed to be highly valued by dominant groups, marginalized peoples and those who listen attentively to them will point out that from the perspective of marginal lives, the dominant accounts are less than maximally objective. Knowledge claims are always socially situated, and the failure by dominant groups critically and systematically to interrogate their advantaged social situation and the effect of such advantages on their beliefs leaves their social situation a scientifically and epistemologically disadvantaged one for generating knowledge." (pg 442)
"Standpoint work must always be "intersectional," in the phrase of the critical race theorists (Crenshaw et al. 1995)." pg 194.
"In much of feminist theory and, to some extent, in antiracist politics, this framework is reflected in the belief that sexism or racism can be meaningfully discussed without paying attention to the lives of those other than the race-, gender- or class-privileged. As a result, both feminist theory and antiracist politics have been organized, in part, around the equation of racism with what happens to the Black middle-class or to Black men, and the equation of sexism with what happens to white women." (Crenshaw, 1989)
"Man has said that woman can be defined, delineated, captured—understood, explained, and diagnosed—to a level of determination never accorded to man himself, who is conceived as a rational animal of free will. Where man's behavior is
underdetermined, free to construct it's own future along the course of its rational choice, woman's nature has overdetermined her behavior, the limits of her intellectual endeavors, and the inevitabilities of her emotional
journey through life." (406)
"Identity politics provides a decisive rejoinder to the generic human thesis and the mainstream methodology of Western political theory. According to the latter, the approach to political theory must be through a "veil of ignorance"
where the theorist's personal interest and needs are hypothetically set aside. The goal is a theory of universal scope to which all ideally rational, disinterested agents would acquiesce if given sufficient information. Stripped of
their peculiarities, these rational agents are considered to be potentially equally persuadable. Identity politics provides a materialistic response to this and, in doing so, sides with Marxist class analysis. The best political
Theory will not be one ascertained through a veil of ignorance, a veil that is impossible to construct. Rather, political theory must base itself on the initial premise that all persons, including the theorist, have fleshy, material
identity that will influence and pass judgment on all political claims. Indeed, the best political theory for the theorist herself will be one that acknowledges this fact. As I see it, the concept of identity politics does not
presuppose a prepackaged set of objective needs or political implications but problematizes the connection of identity and politics and introduces identity as a factor in any political analysis." (432)
"We believe that the most profound and potentially most radical politics come directly out of our own identity"
"A political contribution which we feel we have already made is the expansion of the feminist principle that the personal is political. In our consciousness-raising sessions, for example, we have in many ways gone beyond
white women's revelations because we are dealing with the implications of race and class as well as sex."
"One issue that is of major concern to us and that we have begun to publicly address is racism in the white women's movement. As Black feminists we are made constantly and painfully aware of how little effort white women
have made to understand and combat their racism, which requires among other things that they have a more than superficial comprehension of race, color, and Black history and culture. Eliminating racism in the white women's
movement is by definition work for white women to do, but we will continue to speak to and demand accountability on this issue."
"Linda Alcoff has introduced the idea of positionality, a concept that emphasizes how individuals come to knowledge-making processes from multiple positions, each determined by culture and context. All of these ideas
offer alternatives to the quest for universal objectivity." (83-84)
"The belief that universal objectivity should be our goal is harmful because it’s always only partially put into practice."…."The key to fixing the problem is to acknowledge that all science, indeed all work in the world
is undertaken by individuals. Each person occupies a particular perspective, as Haraway might say; a particular standpoint, as Harding might say; or a particular set of positionalities, as Alcott might say." (83)
"Disclosing your subject position(s) is an important feminist strategy for being transparent about the limits of your—or anyone's—knowledge claims. Thus, for example, we (the authors) included statements about our own
positionalities in the introduction in order to disclose the gender, race/ethnicity, class, ability, education, and other subject positions that informed the writing of this book. Rather than viewing these positionalities
as threats of as influences that might have biased our work, we embraced them as offering a set of valuable perspectives that could frame our work. This is an approach that we would like to see others embrace as well.
Each person's intersecting subject positions are unique, and when applied to data science, they can generate creative and wholly new research questions." (83)
"This goal reflects a key tenet of feminist thinking, which is the recognition that a multiplicity of voices, rather than one single loud or technical or magical one, results in a more complete picture of the issue at hand." (136)
"...all knowledge is partial, meaning no single person can claim an objective view of the capital-T Truth." (32)
"Without contextualising the researcher and research environment in qualitative studies, often the meaning of any research output is lost. What follows is that positionality does not undermine the truth of such research,
instead it defines the boundaries within which the research was produced. The absence of positionality when considered alongside the notion of bias, may challenge the quantitative idea of
validity." pg 323
"Positivism denotes that knowledge comes from objective and rigorous scientific measurement and testing to provide a fixed answer, whereas constructivism denotes that knowledge depends entirely on subjective perception and
consequently is not a fixed entity. Positivism, which fits into the more generic concept of 'hard-science', considers something to be true, false or without meaning. For something to be meaningful, it must
be able to logically be proven or disproven." pg 323
"This absence of positionality does not provide opportunity for the audience to decide how important these factors might be and as a consequence this reduces the validity of the research
conclusions." 323
Paper Site
Regarding using positionality in the design of machine learning systems.
"Positionality, and how it embeds itself in standards, ontologies, and data collection, is the root for bias in our data and algorithms.
Every perspective has its limits - there is no view from nowhere. Without an awareness of positionality, the current debate on bias in machine learning is quite limited: adding more data to the set cannot remove
bias. Instead, we propose positionality-aware ML, a new workflow focused on continuous evaluation and improvement of the fit between the positionality embedded in ML systems and the scenarios within which
it is deployed."
Positionality and reflexivity in regards to the machine learning models and the data scientists who make them, making choices on what to include and how to interpret it based on what they see as important — as influenced by positionality.
Searched engineering education journals for articles with positionality statements and found 15, all qualitative. Includes examples, and includes positionality statements from the authors.
"4.1 Researcher Positionality
Reflexivity in research practice establishes the researcher as a lens through which research is conducted; what Attia and Edge call an “on-going mutual shaping between researcher and research” [13].
In other words, the research is shaped by the positionality—the social and political context—of the researcher. In alignment with the feminist practice of reflexively examining one’s relationship to one’s
research, we want to highlight how the positionality of the authors may have shaped this work.
Our approaches to examining both gender and race are informed by our collective experience— and many other experiences that make up our perspectives as researchers. The second author is Black, while the
remaining three authors are white. Every author has a different gender. All of the authors are based in the United States. As such, our experiences are rooted in a Western-centric point of view. Each author
comes from a multidisciplinary background, including HCI, computer science, psychology, gender studies, communication, and the arts. Our synthesized experiences with critical theory, race studies, and gender
studies are shaped by education (both formal and informal) related to our U.S. nationalities. Our decision to examine computer vision practice with a critical lens stems from our scholarly upbringings.
Our privilege as academics awarded us access to the resources to conduct this work, while our power as differentially marginalized individuals gave us the perspective to develop our research questions
and interpret our data."
"The imagined "they" constitute a kind of invisible conspiracy of masculinist scientists and philosophers replete with grants and laboratories. The imagined "we" are the embodied others, who are not allowed not to have a body, a finite point of view, and so an inevitably disqualifying and polluting bias in any discussion of consequence outside our own little circles, where a "mass" subscription journal might reach a few thousand readers composed mostly of science haters." (575)
"It is important to note we are not advocating for the compulsory disclosure of sensitive experiences or marginalized identities. We acknowledge the increased burden of researchers from marginalized identities in self-disclosure, which may risk their personal and professional lives [92]. There is also the valid concern that work conducted by marginalized individuals on topics of identity will be viewed as less scientific and less valid [68, 152]. Rather, we are advocating for increased context setting around the decisions researchers make when constructing and documenting databases, and a deeper attention to documenting the identities of annotators as seen in more research practice (e.g., the reporting of participant demographics)."
How machine learning works, and how bias can be embedded via data, and the complications in limiting bias against all groups.
Review of image databases used in facial recognition to see how race and gender are classified.
"Beyond the general lack of source material and annotation descriptions in databases, we also observed a lack of acknowledgment of external identification as a subjective process, informed by one's own position and
perspective in shaping race and gender categories. Without understanding the position of the author or annotator, the collapse of subject identity is made to appear neutral or objective. Statements like "most ethnicities
and races were included" (CFEE [39]) and "the diversity ... is guaranteed by the large scale of our dataset" (MS-CELEB-1M [59]) are written into database documentation as if the comprehensive diversity of
race is objectively possible." 58:21
"As race and gender are sociohistorically situated, so too are the perceptions authors and annotators introduce into databases. Including the perspectives, training, and identities authors and annotators bring
to image databases would increase the level of transparency currently absent in decisionmaking processes. Knowing the demographic distribution of authors and annotators is just as useful as knowing
the demographic distribution of subjects in the database." 58:24
You can download the dataset and see information on the source of the poll and the results from each source, but no information on how it was collected or who it was collected from — without that context there is no way to generalize this to any population. You can click through to each individual data source to find more information (if it exists there), but on 538 in aggregate, there is not context to the data. Political data is highly contextual.
Alcoff, L. (1988). Cultural Feminism versus Post-Structuralism: The Identity Crisis in Feminist Theory. Signs, 13(3), 405–436.
Boté, J.J., & Termens, M. (2019). Reusing Data Technical and Ethical Challenges. DESIDOC Journal of Library & Information Technology, 39(06), 329–337. https://doi.org/10.14429/djlit.39.06.14807
Bruynseels, K. (2020). Responsible innovation in synthetic biology in response to COVID-19: The role of data positionality. Ethics Inf Technol, 26(23(Suppl 1)), 1–9. https://doi.org/doi:10.1007/s10676-020-09565-9
Combahee River Collective. (2012, November 16). (1977) The Combahee River Collective Statement. https://www.blackpast.org/african-american-history/combahee-river-collective-statement-1977/
Crenshaw, K. (1989). Demarginalizing the Intersection of Race and Sex: A Black Feminist Critique of Antidiscrimination Doctrine, Feminist Theory and Antiracist Politics. The University of Chicago Legal Forum, 140, 139–167.
Data Documentation–Dimewiki. (n.d.). Retrieved October 9, 2022, from https://dimewiki.worldbank.org/Data_Documentation
Dean, J., Furness, P., Spencer, S., Verrier, D., Lennon, H., & Bennett, C. (2018). Desert island data: An investigation into researcher positionality. 2018. Qualitative Research, 18(3), 273–289. https://doi.org/10.1177/1468794117714612
D'Ignazio, C., & Klein, L. F. (2020). Data Feminism. https://doi.org/10.7551/mitpress/11805.001.0001
Ferreira, D., & Vale, M. (2022). GEOGRAPHY IN THE BIG DATA AGE: AN OVERVIEW OF THE HISTORICAL RESONANCE OF CURRENT DEBATES. Geographical Review, 112(2), 250–266. https://doi.org/10.1080/00167428.2020.1832424
Haraway, D. (1988). Situated Knowledges: The Science Question in Feminism and the Privilege of Partial Perspective. Feminist Studies, 14(3), 575–599. https://doi.org/10.2307/3178066
Harding, S. (1986). The science question in feminism. Ithaca: Cornell University Press, 1986. https://www.cornellpress.cornell.edu/book/9780801418808/the-science-question-in-feminism/#bookTabs=1
Harding, S. (1992). RETHINKING STANDPOINT EPISTEMOLOGY: WHAT IS “STRONG OBJECTIVITY?” The Centennial Review, 36(3), 437–470.
Harding, S. (2009). Standpoint Theories: Productively Controversial. Hypatia, 24(4), 192–200.
Heilweil, R. (2020, February 18). Why algorithms can be racist and sexist. Vox. https://www.vox.com/recode/2020/2/18/21121286/algorithms-bias-discrimination-facial-recognition-transparency
Jafar, A. J. N. (2018). What is positionality and should it be expressed in quantitative studies? Emergency Medicine Journal, 35(5), 323–324. https://doi.org/10.1136/emermed-2017-207158
Kitchin, R. (2014). Big Data, new epistemologies and paradigm shifts. Big Data & Society, 1–12. https://doi.org/10.1177/2053951714528481
Liabo, K., Cockcroft, K. B., Farmer, L., Bortoli, S., & Britten, N. (2022). Epistemic justice in public involvement and engagement: Creating conditions for impact. Health Expectations, 25(4), 1967–1978. https://doi.org/10.1111/hex.13553
Olukiun, O., Mkandawire, E., Antilla, J., Alfaifa, F., Weitzel, J., Scheer, V., Olukotun, M., & Mkandawire-Valhmu, L. (2021). An Analysis of Reflections on Researcher Positionality. The Qualitative Report. https://doi.org/10.46743/2160-3715/2021.4613
Positionality. (2014). In D. Coghlan & M. Brydon-Miller, The SAGE Encyclopedia of Action Research. SAGE Publications Ltd. https://doi.org/10.4135/9781446294406.n277
Scheuerman, M. K., Wade, K., Lustig, C., & Brubaker, J. R. (2020). How We've Taught Algorithms to See Identity: Constructing Race and Gender in Image Databases for Facial Analysis. Proceedings of the ACM on Human-Computer Interaction, 4(CSCW1), 1–35. https://doi.org/10.1145/3392866
Wiederkehr, R. B., Aaron Bycoffe, Ritchie King, Dhrumil Mehta and Anna. (2018, June 28). Generic ballot Polls. FiveThirtyEight. https://projects.fivethirtyeight.com/polls/