Literature Review Tips for Social Science PhD Students

Literature reviews still matter. AI tools now do much of the work of finding papers, summarizing them, and even drafting a review, but the review is still the section where a researcher shows a sharp view of the field: which findings rest on weak measurement, for example, and where a new study has room. Boote and Beile, writing about doctoral training in education, argued that the dissertation literature review is the foundation for substantial, useful research, and they proposed criteria for judging its quality after examining dissertations at three universities.

Citation and engagement

Citing a paper and engaging with it are different acts. A citation names a paper as support for a sentence. Engagement says what the paper did, how it did it, what it leaves open, and where your study stands relative to it.

In the made-up example below, the first version only cites a source.

> Language barriers are associated with lower mental health service use among immigrants (Study A, > 2019; Study B, 2021).

The second version engages with the two studies.

> Study A (2019) and Study B (2021) both found lower mental health service use among immigrants > with limited English proficiency, but both measured language barriers with a single self-rated > English item and could not distinguish the person’s own proficiency from the availability of > providers who speak their language.

The second sentence is longer, and it does more work. A reader now knows what the two studies measured, where they agree, and which question they could not answer, and that last part is what makes room for your study. The Purdue OWL guide describes a literature review as a discussion of sources in conversation with each other, and that is a useful picture: the conversation started before you arrived, so the review reports the points others have made and then adds yours.

Research lineages

Before writing a single paragraph, I try to sketch the family tree of the literature: which studies the field builds on, which lines of work branched from them, and how those lines differ in their questions, data, and explanations. Immigrant health research, for example, has a long line of work explaining outcomes through acculturation and a newer line focused on discrimination, place, and immigration policy, and a study reads differently depending on which branch it extends. AI tools and citation graphs can now find most of the papers that belong in this tree, but drawing it, on paper or in a matrix, is your own work, because where the branches divide depends on the position you view the field from: your discipline, your training, and the values you bring to the problem.

Several clues help in drawing it. Research groups are one: a lab or a set of collaborators often continues a line of work across many papers, so the coauthor lists and the people a group trained show where a branch starts and how it spreads. The social network of a field works the same way, since scholars who publish together, cite each other, and meet at the same conferences tend to share assumptions about what the problem is. Keywords are a third clue: two lines of work studying the same population often use different vocabularies, such as acculturation and assimilation in one and structural racism and immigration policy in the other, and searching with only one vocabulary returns only one branch.

Disciplinary training is a fourth clue, so for each theory, school, or claim in the tree I recommend checking which fields its scholars come from and what training they received, because scholars in different fields explain the same topic differently. On immigrant health, a political scientist may look to immigration policy and political incorporation, a policy scholar to program rules and eligibility, a sociologist to assimilation and stratification, a social psychologist to identity, stereotyping, and acculturative stress, and a social work scholar to access to services and practice. Explanations also divide within a field by the scholar’s theoretical and philosophical commitments. A post-structuralist account asks how categories such as “immigrant” or “limited English proficient” are produced in policy and clinical discourse, while a network approach studies the ties through which information and support pass between people. I described the same pattern in computational social science, which looks like one field from outside but divides into communities with different training, data, and venues.

Choosing a method or a theory often also states a value. Caplan and Nelson (1973) argued that psychological research on social problems tended to locate their causes in individuals, which pointed toward changing people instead of the conditions around them, and Rappaport (1981) contrasted programs that experts design for a community with solutions built on the community’s own capacities. Whether a study approaches a problem as a multilevel question or as a psychological process within individuals, and whether it looks for solutions inside a community or outside it, says something about what the researcher believes causes the problem and who should act on it. Crotty describes the research process as a chain from epistemology to theoretical perspective to methodology and methods, so a choice of method rests on philosophical commitments whether or not the study names them.

Mills asked social scientists to connect individual lives to the history of the society around them, and it helps to think about a school of research historically in the same way, since its claims make more sense once you know when it formed, who trained in it, and what earlier explanation it was responding to. The position from which you draw the branches is also the one you are building as a researcher, so noticing which of these commitments you agree with is part of how your research identity forms during the doctoral program.

A few free tools make these clues visible:

• Google Scholar author profiles list frequent coauthors, which is a quick way to see who works with whom.
• OpenAlex is an open catalog of works, authors, institutions, and topics, so you can pull every paper by a group or on a topic and see who cites whom.
• VOSviewer draws network visualizations of bibliographic data, including co-authorship and keyword co-occurrence networks, from a set of exported records.
• Connected Papers and ResearchRabbit start from one paper and show the papers around it, which helps when you know a single study and not yet the branch it belongs to.

Annual Reviews articles, such as those in the Annual Review of Sociology or the Annual Review of Public Health, help with this. Authors invited to write them describe how a field developed and group its work into lines of research, so reading one before you draft gives you a structure to compare with your own tree. Where their grouping and yours differ, ask why; the difference often points to the question your study can address. Connected Papers shows a similar tree from citation data through its prior and derivative works views.

Claims as the organizing unit

A review that goes “Author A found X. Author B found Y. Author C found Z.” is an annotated bibliography. Pat Thomson calls this the laundry list literature review, and her tell is easy to check in your own draft: most sentences start with a scholar’s name. Her rewrite makes the paragraph’s claim the subject of the sentence and moves the names into the citations. A review is organized by claims or themes, and each paragraph makes one point with several papers behind it. When I see a paragraph built on a single source, it is usually a summary of that source, and I check whether the paper belongs inside a larger point.

Webster and Watson called this the difference between an author-centric and a concept-centric review, and they suggested a concept matrix to organize it: papers as rows, concepts as columns. With a complete matrix, a paragraph often follows from a single column: three studies agree on the direction of an association, one reports the opposite, and the difference lines up with how each study defined the sample.

The UNC Writing Center’s handout on literature reviews adds a habit I ask students to copy. Start and end each paragraph with your own idea in your own words, and let the sources support it in between, so your own argument leads and the citations follow.

Synthesis sentences tend to have a few recognizable shapes, and I use these the most:

• Agreement. Several studies found the same pattern across different samples or designs, so the finding does not depend on one sample.
• Disagreement. Studies arrived at different conclusions, and the difference may come from measurement, population, or design.
• Conditions. An association appeared only in some settings or for some groups.
• Silence. A question none of the studies could address with the data or design they had.

Classic and recent studies

A review built only on foundational work tells the reader that the author may not know the current field. Early drafts often build the framing on a theory from the 1980s and a few widely cited studies from the 2000s, with nothing from the last five years. From that list, a reviewer has no way to judge whether the question is still open or whether someone answered it in 2023.

The first check concerns the newest citation. Before sending a draft to anyone, I look at the most recent year cited in the introduction and the discussion, and if it is more than three to five years old, I search again. The second check is pairing. For each foundational source, I ask whether recent work has tested, extended, applied, or contested it, and I cite that work next to the classic. The classic gives the origin of the idea, and the recent study explains where the field is now.

Closest prior studies

The paragraph that states your contribution is the one reviewers read most closely, and in my experience it is the one most likely to be vague. “Few studies have examined this topic” invites the reply that the reviewer has read three of them.

I recommend naming the two or three studies closest to yours and saying, for each one, what it did and what it did not do that your study does. Written out, it looks something like this made-up example: Study C (2022) used the same national survey but pooled all immigrants into one group, and Study D (2024) disaggregated by language but relied on a single state. The present study combines national coverage with language-level estimates.

The gap usually falls into one of these types:

• Population gap. Prior studies did not include a group, or pooled it with others.
• Measurement gap. The construct was measured coarsely, for example with one item where a validated scale exists.
• Methodological gap. The design could not support the claim being made, for example causal language from cross-sectional data.

Gaps are not the only way to make room for a study. Locke and Golden-Biddle examined how published articles construct opportunities for contribution, and their title names two moves: building coherence across prior work and problematizing it. Sandberg and Alvesson reviewed 52 articles in organization studies and found that spotting gaps was the most common way researchers formulated their questions. In a companion paper, Alvesson and Sandberg proposed problematization, identifying and challenging the assumptions under existing work, as the alternative. Pat Thomson makes a related point in a post on blind spots: the definitions, categories, and survey items you borrow from the literature can bring its assumptions into your own study.

A gap claim is only as good as the search behind it. Before writing that something has not been studied, I search again with different terms, check the citing articles of the closest prior study, and enter the closest study into a tool such as Connected Papers or ResearchRabbit, since both find related work that keyword searches miss.

Examples

The four articles below come from health services, public health, and immigrant health research, and each frames its study against prior work in a way you can follow sentence by sentence.

• Sentell, Shumway, and Snowden (2007). Access to Mental Health Treatment by English Language Proficiency and Race/Ethnicity, Journal of General Internal Medicine. The introduction notes a common assumption that limited English proficiency drives mental health disparities, with little research testing it directly. It then names what the closest prior studies could not do: one study included only insured patients, and studies of nativity and acculturation could not separate those factors from English proficiency because they lacked a proficiency measure.
• Karliner, Jacobs, Chen, and Mutha (2007). Do Professional Interpreters Improve Clinical Care for Patients with Limited English Proficiency?, Health Services Research. The introduction builds on earlier reviews, including Flores (2005), and notes that many of the studies behind them combined ad hoc interpreters, trained interpreters, and bilingual clinicians. The specific effect of professional interpreters is the question left open, and the review is designed around it.
• Abe-Kim and colleagues (2007). Use of Mental Health-Related Services Among Immigrant and US-Born Asian Americans, American Journal of Public Health. The introduction is organized by claims: mixed findings on immigration-related factors, scarce data on satisfaction with care, and barriers grouped into cultural, service, access, and awareness categories. It closes by naming the limits of prior samples (treated patients, college students, and convenience samples) and their narrow range of disorders, which the national sample was designed to address.
• Viruell-Fuentes, Miranda, and Abdulrahim (2012). More than culture: Structural racism, intersectionality theory, and immigrant health, Social Science & Medicine. This is a review article, and its abstract describes problematization at the scale of a field: it questions the reliance on acculturation as the explanation for immigrant health outcomes and proposes lines of inquiry on discrimination, place, and immigration policy.

Notes for the review

Most of this work happens at the note-taking stage, well before the writing. These habits matter most for the review itself:

• Read with a purpose. A review paper, a methods reference, and a paper you plan to position against each call for a different depth of reading.
• Paraphrase in two or three sentences. A note in your own words about what the paper did and how it relates to your study is useful at the writing stage in a way a highlighted PDF is not.
• Record what the paper did well and what it could not do. A strong design, a validated measure, or a clear way of reporting results is something your study can build on, and the limitation you note while reading often becomes the sentence that positions your study.
• Use a reference manager from the start. I use Paperpile, and Zotero is a free option. Either one saves hours when a journal asks for a different citation style.

Guides to writing the review section

These are the guides I send students to, with what each one covers:

• UNC Writing Center, Literature Reviews: a handout on chronological, thematic, and methodological organization, with advice on using evidence, selecting points, quoting sparingly, and keeping your own voice.
• Purdue OWL, Writing a Literature Review: a guide to where a review appears in a paper, how disciplines differ, and the parts of a review.
• USC Libraries, The Literature Review: a chapter of the Organizing Your Social Sciences Research Paper guide on summary versus synthesis and the ways reviews are organized.
• Pat Thomson, avoiding the laundry list literature review and her literature review category: blog posts on doctoral writing, including rewrites of name-first paragraphs.
• Grant and Booth (2009), A typology of reviews: an article describing 14 review types, for when a standalone review such as a scoping or systematic review is the goal.

Final checks

Before I send a review section to a coauthor, I reread it with these questions:

• Does each paragraph make one point, with more than one source behind it?
• Do most sentences start with the point, with the names in the citations?
• Is something from the last three to five years cited, and is each classic paired with recent work?
• Are the closest prior studies named, with what each did and did not do?
• Does every gap claim rest on a search I would be comfortable describing to a reviewer?
• Could a reader tell, from the review alone, why this study comes next?