From Spam to Impact: A Practical Workflow for Writing Psychology Papers Based on Secondary Data

1) Start With an Insight and Map the Literature

Ever sat in your email reading list thinking: “I know there’s a pattern here, but where’s the actual research?” That’s where secondary data analysis shines. You take the insight living rent-free in your head—we’re going to use the inadequacy–dehumanization connection as an example—and you turn it into publishable psychology by piggybacking on high-quality existing datasets. Thats how you gain traction writing a psychology paper and scientific papers in general.

Here’s the honest truth: Secondary analysis is one of the most efficient ways to do serious psychological research when you don’t have the budget, time, or ethics clearance for large primary studies. It lets you add your own conceptual spin, move quickly from idea to manuscript, and still respect rigor, ethics, and transparency.

The whole process starts with an insight you actually care about, ideally rooted in both your own experience and your reading.

Mapping the State of the Art

Once you have that seed for writing your psychology paper, your first task is to map what’s already out there:

Search for primary and secondary studies on the key constructs you care about, even if they’re older but foundational (e.g., classic SDO scales, dehumanization measures, stigma and mental health outcomes).
Track both “big picture” theoretical pieces (e.g., the vicious cycle of dehumanization and societal tension) and highly operational work (specific scales, demographics, geographies, and measurement strategies).
Note down what has been consistently measured (e.g., SDO, perceived inadequacy, contempt, mental health symptoms), in which populations, and in which countries or cultures.

At this point you want a living document—your “spam folder” for ideas—listing: constructs, measurement instruments, known demographic and geographic trends, and any recurring mediators or moderators (e.g., echo-chamber usage, social media intensity).

AI “Macro Prompt” for Scoping the Topic

Once you have your preliminary basis, use an AI assistant to synthesise and structure the topic space. Drop your own summary of the insight and your reading into a conversation, then use this prompt (adjust the part in square brackets to your research topics):

I would like you to create a comprehensive, in-depth guide to research into the above that covers every important stage and element of the process from start to finish. Include information on [The Inadequacy-Dehumanization Connection, Social Dominance Orientation (SDO), The Role of Contempt in Dehumanization, Echo Chambers and the Amplification of Dehumanization, The Vicious Cycle of Dehumanization, and The Implications for Mental Health and Societal Tensions]. What measurements are possible? What trends are there in demographics and Geographies? Include Quantitative and Qualitative as well as mixed methods Research, data collection methods: Surveys and Questionnaires, Observations, Existing Data, data analysis types: Descriptive Statistics, Inferential Statistics, Multivariate Analysis, Content Analysis, Thematic Analysis, results interpretation – use Statistical Significance, Effect Size, Comparison with Hypotheses and Integration of Findings, Discussion Approach – Use both Quantitative and Qualitative as well as mixed methods discussion approaches, conclusion – use approaches: Summarizing Key Findings, Statistical Significance and Impact, Broader Implications, Synthesis of Themes, Contributions to Theory, Narrative Closure and Reflective Insights, writing and formatting: Should be written acting as an academic using research paper structured format, Include peer review methods, Include Iterative revision approach, Use proofreading, Describe submission process and methods, publication approaches, dissemination: provide options for wide distribution, and how to measure impact tracking.

Explain the significance of each stage, outlining best practices, potential challenges, and strategies for successful execution. Emphasize the iterative nature of the research process, the importance of rigor and ethical considerations, and the role of collaboration and feedback. Address common pitfalls and offer tips for enhancing the quality and impact of research papers.

Answer the questions. Use the AI output as a structured checklist, not as a replacement for reading the actual papers.

2) Formulate Research Questions and Hypotheses

The next job is to turn your insight and literature map into specific research questions (RQs) and hypotheses. These are the spine of your paper: they determine which datasets you select, what you code, and how you analyse.

Prompt Well: Research Questions and Hypotheses

Feed your insight and key theories into an AI conversation, then use this prompt:

Detail the process of formulating research questions and hypotheses in research papers. Explain the criteria for creating clear and related questions, and discuss how hypotheses guide the research design and contribute to a structured approach.

Then refine the AI’s answer into your own RQs and hypotheses.

Criteria for Good Research Questions

When you edit the AI’s suggestions, enforce a few strict criteria:

Each RQ should be clear, focused, and answerable with the variables you can actually get from the secondary datasets below.
RQs should be logically related to each other, often moving from description (What is the prevalence of X?) to explanation (Is X associated with Y?) to moderation/mediation (Does Z change or explain the X–Y link?).
RQs must be grounded in the theory you have and want to explain in the paper, not just what happens to be in the dataset; otherwise the paper becomes a fishing expedition.

How Hypotheses Guide Design

Hypotheses are testable statements that predict the direction or presence of relationships between variables and they get proven or disproven to form the paper, they directly guide:

Which variables you extract or derive from a dataset (e.g., SDO scores, self-reported inadequacy, dehumanizing language scales, mental health indices).
Which analytic approaches you use (e.g., correlation, regression, mediation models, multivariate analysis).
How you structure your Methods and Results (each hypothesis should have clear analytic tests and a clearly labelled result subsection).

A structured set of hypotheses also makes your paper easier to peer review, because reviewers can trace each claim back to a specific test and effect size.

Example Set of 5–7 Hypotheses

For a dehumanization-centred project that might generate multiple papers, you might pre-specify hypotheses like:

Higher self-reported inadequacy will be positively associated with dehumanizing attitudes toward out-groups, controlling for age, gender, and education.
SDO will partially mediate the relationship between inadequacy and dehumanizing attitudes.
Contempt toward out-groups will predict greater endorsement of dehumanizing metaphors, beyond general negative affect.
Time spent in ideologically homogeneous online spaces (echo chambers) will strengthen the association between inadequacy and dehumanizing attitudes (a moderation effect).
Higher dehumanization scores will be associated with poorer self-reported mental health and higher societal tension indicators (e.g., perceived conflict, threat).
Null-type hypothesis: After controlling for SDO and contempt, inadequacy will no longer have a unique association with dehumanization in some datasets.
Null-type hypothesis: Echo-chamber exposure will not significantly differ in its association with dehumanization across countries with high versus low polarization scores in some samples.

3) Source Relevant Secondary Data

Searching for sources to back your knowledge while writing your psychology paper is the foundation of a good paper. Good data makes the paper high quality. With hypotheses in hand, you now look for data that can plausibly test them. Secondary datasets for psychology and social science are increasingly easy to find through institutional repositories, government surveys, and curated lists.

You can use this list of sources and search engines to find out which datasets might help your study: Awesome Psychology and Social Science Data by HomingHamster on Github.

Make sure you credit the actual datasets according to the licensing terms, many allow free use in publications if correctly cited. In parallel, university library guides to psychology secondary datasets and national statistics portals can help you locate large, representative samples.

For each potential dataset, check:

Does it contain constructs that map reasonably onto your variables (e.g., SDO scale, prejudice measures, mental health questionnaires, online behaviour items)?
Are dehumanization-adjacent items present (e.g., metaphor items like “animals,” “vermin,” or “savages,” or more subtle humanness denials)?
What demographics and geographies are covered (age ranges, gender, country, region, urban/rural, industry)?
Are there enough cases and enough variance in your key variables to detect the effects you care about with adequate power?
Does the combined data act to either prove or disprove all the hypotheses?

Ethically, remember that even though the data already exist, you may still need an ethics review or at least an exemption letter from your institution, particularly if the data are sensitive.

Optional: Pre-processing With Coding Schemes and AI

Once you have the raw data, you may need to preprocess it, perhaps by adding columns deriving dehumanization scores from text fields or adding multi-item scales, for example, to create a new dataset. You can:

Create coding schemes that classify text into categories (e.g., high vs low dehumanizing language; contempt-laden vs neutral; in-group vs out-group target).
Use AI models as coding assistants to label text snippets according to your scheme, with human spot-checking for reliability.
Compute scale scores, reverse-code items, handle missing data, and create composite indices (e.g., an echo-chamber exposure index from multiple related items).

Document all preprocessing steps when writing your psychology paper; in secondary data analysis, transparency about how you transformed the original variables is crucial for credibility. This is part of your method section.

4) (Optional) Primary Source Questionnaire Design to Complement Secondary Data

Even in a secondary-data-first project, you may want to design a short new questionnaire to validate your constructs or to replicate patterns in a different sample. This is especially useful if you suspect measurement limitations in the archival datasets.

Prompt Well: Questionnaire Design

Feed the AI your hypotheses and a summary of the secondary data variables you have, then use this prompt:

Design a questionnaire according to the description and hypotheses above. Include detailed informed consent steps explaining the risks and the mitigations, the study will not relate identifiable data to the study questionnaire answers. Collect the data needed to reliably categorise the response into the necessary categories, such as the country, the industry. Design questions that confirm and verify existing data. Ask important questions in different ways. Make sure there is a test for every hypothesis in multiple ways, use a mixture of open and closed questions, suggest metrics that can be used to analyse each question to prove or disprove the hypothesis. At the end of the questionnaire, gain permission to store identifiable data which will only be used to enter the participant into a prize draw for one of 7 $10 Amazon gift cards. Use language a 9 year old would understand without being condescending. The questionnaire should be detailed and take the user around 15 minutes to half an hour.

Then refine: remove leading questions, simplify wording, and ensure informed consent clearly covers anonymity, risks, and data use. Make sure every hypothesis is probed in more than one way (e.g., multiple items for inadequacy; several dehumanization items; parallel questions about echo-chamber use) so you have robust measures and can check internal consistency.

You can use Formbricks to present the form on your own server or paid solutions like Google Forms.

5) Profile the Data With ydata-profiling and scikit-learn

This is one of the most interesting steps in writing your psychology paper and gives you the evidence for your insight. Once your secondary (and optional primary) data are cleaned, you want an overview of distributions, missingness, and correlations. Tools like ydata-profiling (formerly pandas-profiling) can automatically generate detailed HTML reports that summarize each column and highlight potential issues.

A minimal example in Python:

import pandas as pd
from ydata_profiling import ProfileReport  # ydata-profiling

df = pd.read_csv("dehumanization_dataset.csv")

profile = ProfileReport(
    df,
    title="Dehumanization & SDO Profiling Report",
    explorative=True
)
profile.to_file("profiling_report.html")

By default, ydata-profiling computes an “auto” correlation matrix that chooses appropriate correlation statistics depending on variable types (e.g., Spearman for numeric–numeric, Cramér’s V for categorical–categorical, discretised associations for mixed types). This helps you quickly see which variables are strongly associated and which may be redundant or highly collinear.

You can then use scikit-learn or standard statistical packages for more focused modelling, always keeping an eye on multicollinearity and overfitting. For example, you might:

Run logistic or linear regressions to test whether inadequacy predicts dehumanization scores while controlling for demographics.
Fit mediation or moderation models (e.g., inadequacy → SDO → dehumanization; echo-chamber use as moderator) using appropriate libraries or SEM tools.
Do dimension reduction (e.g., PCA) to construct composite indices from correlated items if conceptually justified.

The correlations you spot and graphs and tables from these steps (correlation heatmaps, distribution plots, partial regression plots) are what you will later convert into LaTeX figures (again using AI).

6) Select Graphs and Note the Key Correlations

From the profiling and modelling outputs, you now curate a small set of figures that best explain your core insight. Resist the temptation to dump every chart into your paper; instead, choose those that directly illuminate your hypotheses and talk about the interesting ones breifly.

Typical candidates example:

A correlation heatmap showing the relationships among inadequacy, SDO, contempt, dehumanization scores, and mental health measures, annotated with the statistic used (e.g., Pearson’s r or Spearman’s ρ).
A scatterplot (with regression line) of inadequacy vs dehumanization, possibly faceted by echo-chamber exposure level.
A bar or violin plot comparing dehumanization by SDO tertile or by geographic region or industry.

For each selected figure, write a one-line memo to yourself: which correlation or effect does this visualize, which hypothesis does it test, and what is the key takeaway to mention in the Results text. These notes make later drafting much faster and more coherent.

7) Draft Intro, Method, Results, and Discussion, With AI as a Co-Author

This is the most important step where you are actually are actually writing the psychology paper. You now have: insight, literature map, RQs and hypotheses, datasets, preprocessing decisions, and analytical results. Time to turn that into a full paper structure, aligned broadly with APA or journal norms: Introduction, Method, Results, Discussion, plus references.

Most APA-style psychology papers follow this structure, even when they are based on secondary data rather than primary experiments.

Introduction: Sets up the conceptual story, reviews key literature (inadequacy, SDO, dehumanization, contempt, echo chambers, mental health), and ends with your RQs and hypotheses.
Method: Describes datasets, participants, measures, preprocessing, and analytic strategy, including how you used secondary data and any new questionnaires.
Results: Reports descriptive statistics, model outputs, and figure references without over-interpreting.
Discussion: Interprets findings, synthesises quantitative and qualitative evidence, outlines implications and limitations, and suggests future work.

You can feed your structured notes and a bullet-point outline into an AI and ask it to draft each section, making clear that you will revise for accuracy and style and that you need explicit placeholders for citations. When you get AI drafts back:

Check every factual statement against your data and your sources; AI is a drafting tool, not an authority.
Ensure each hypothesis is clearly mapped to specific analyses and results in the text.
Add explicit mention of ethical approvals and data-use permissions in the Method section, especially for secondary datasets.

Use iterative passes: first content, then clarity and structure, then style and APA formatting.

8) Write the Abstract and Keywords

Only once the full paper body is reasonably stable should you write the abstract. In psychology, an abstract is typically 150–250 words summarising the whole study including background, methods, key results, and implications.

A good abstract for this kind of project will:

Name your key constructs (e.g., inadequacy, SDO, dehumanization, contempt, echo-chamber exposure, mental health).
Briefly describe your use of secondary data, sample characteristics, and analytic methods (e.g., multivariate regression, mediation analysis).
Highlight statistically and practically meaningful findings, including effect sizes.
Mention implications for mental health and societal tensions.

You can ask an AI to propose several candidate abstracts, then edit the best one for precision and concision. Afterwards, choose 3–5 keywords (e.g., “dehumanization,” “social dominance orientation,” “secondary data analysis,” “echo chambers,” “mental health”) and list them on a new line after the abstract if the target journal uses keywords.

9) Convert to LaTeX and Integrate Figures and References

Most of the work is done, this step in writing your psychology paper is about making it look good. Once your draft is stable in a word processor or markdown, you can convert it to LaTeX using an AI assistant or a converter and then hand-edit. This will make it look amazing. LaTeX is still the preferred format for many social-science and interdisciplinary journals, and it makes managing figures and references much more robust.

During this stage we will convert the text into latex formatting that we can paste into TeXWorks. After this we will convert the graphs into LaTeX formatting using AI by adding the data to the context, and asking for a suitable TikZ graphic. Finally, we will make sure the datasets are cited and the past insight bases that we rely on throughout the document are correctly citing the relevant articles, books, and papers. LaTeX compliation can be fustrating, but you can get a long way by googling the errors, and keeping it neat from the start. Sometimes you just need to complie multiple times in a certain order using the right LaTeX compilers for the job.

LaTeX Structure

Your LaTeX document will roughly follow:

Preamble with document class (possibly a journal’s template), packages for graphics, bibliography, and APA style.
\begin{document} with title, author names and affiliations, abstract, and keywords.
Sections: Introduction, Method, Results, Discussion, References, and optionally Appendices.

You can paste the AI-generated content into the appropriate sections, then fix line breaks, escaped characters, and LaTeX-specific syntax.

Converting Graphs to LaTeX Figures

For each selected graph:

Export the plot from Python or R as a high-resolution PDF, PNG, or EPS.
In LaTeX, create a figure environment with \includegraphics, a descriptive caption, and a label (e.g., \label{fig:heatmap_dehum}).
Make sure each figure is cited in the Results section (e.g., “see Figure 1”).

If you want to go further, you can use tools like TikZ or PGFPlots to recreate the graphs directly in LaTeX, but exporting from your statistical environment is usually sufficient and more reproducible.

Building a `references.bib` File

Next, construct a BibTeX file (references.bib) containing all cited works: data sources, measurement instruments, theoretical papers, and methodological references. Use official citation information (DOIs, journal names, volume and issue, page ranges) from publisher pages or indexing services.

For each paper, create a BibTeX entry (@article, @book, @misc, etc.) with all required fields.
Include entries for your secondary datasets, following the citation recommended by the repository or data provider where possible.
Test the bibliography compilation locally to ensure all references appear correctly and that there are no missing fields or duplicate keys.

AI can help generate BibTeX entries, but you must verify that the DOIs, titles, and URLs are real and accurately formatted, because it is highly typical for AI to invent references.

10) Apply APA or Journal Formatting, Credits, Licensing, and Final Checks

Most psychology journals follow APA 7th edition style, either directly or with minor adaptations, so aligning your manuscript to APA by default is a safe baseline. This process involves adding the style tags to the top of the LaTeX document. If you have a working document, journals can often help you apply the correct styling and there are some guides online. At this stage of writing your psychology paper you will reach the point where it looks journal ready.

APA and Journal Formatting

Key APA 7th edition elements include:

Title page with title, authors, affiliations, and author note as required.
Consistent headings hierarchy, double spacing, 1-inch margins, and appropriate font (e.g., 12-point Times New Roman or equivalent).
In-text citations and reference list formatted according to APA rules (author–date style, use of “et al.,” DOI formatting).
Standard paper structure (Introduction, Method, Results, Discussion, References), plus tables and figures formatted and labelled per APA.

Templates and annotated sample papers from APA and university writing centres can be invaluable checkpoints at this stage.

Credits and Licensing

For a project rooted in openness and access, consider:

Explicitly crediting data creators and repositories in both the Method and the References, not just with a citation but also acknowledging their role in enabling your secondary analysis.
Choosing a license for your own article preprint (e.g., a Creative Commons license) that matches your goals for reuse and derivative works.
Adding your own credit line on the preprint and on your personal site (e.g., circuspam.coffee), clarifying authorship, contributions, and funding.

If your analysis code and derived datasets can be shared, posting them in an open repository with a clear license further strengthens transparency and impact.

Final Quality, Ethics, and Impact Checks

Before submitting:

Run a meticulous proofreading pass for grammar, clarity, and consistency of terminology while writing your psychology paper (e.g., always using the same labels for constructs like inadequacy and dehumanization).
Check that every table and figure is referenced in the text, and that every citation in the text appears in the reference list and vice versa.
Confirm that your description of ethical approvals and data handling matches institutional and journal expectations for secondary data analysis.

Finally, think about dissemination and impact tracking:

Post a preprint in an appropriate repository, share summaries on platforms your audience actually uses, and link back to the full paper from circuspam.coffee.
Track citations, downloads, and alternative metrics over time, and consider follow-up posts that unpack methods or findings for non-specialist readers.

By treating this entire workflow as iterative—refining hypotheses while writing your psychology paper as you learn more about the data, revising drafts based on peer feedback, and updating analyses as better methods emerge—you turn what might look like noisy “spam” in your conceptual inbox into a growing archive of rigorous, reusable psychological science.

UPDATE: See our article about using Semantic Evaluators to process datasets ready to apply the statistical analysis mentioned in this article.