Skip to content
SaidGig
Sign up.

Give me your email, I promise I won't do anything weird with it.

Article

Which Social-Science Research Approach Fits Your Question?

Read it normal, or read it

Powered by Boligrafun.com

The first step in social-science research is not opening a spreadsheet and staring at it like you are trying to intimidate the numbers. The first step is deciding what you actually want to know.

This sounds obvious. It is not. A question such as “What do employees think about remote work?” may secretly contain several different questions wearing one large raincoat. How many employees prefer it? How do they describe it? Why do they prefer it? What did another researcher’s employees say three years ago? These questions are related, but they are not the same creature. If you feed them all into one survey with 78 questions and a tiny box marked OTHER, the creature becomes angry.

So begin by writing the question plainly. Then inspect its main verb.

Are you trying to count, compare, describe, interpret, or review what is already known? That choice will not mechanically select a perfect method. There is no universally best method sitting on a marble chair. It does, however, reveal what kind of evidence you need.

If the question calls for structured comparisons or numerical analysis, a quantitative approach may fit. If it asks how people describe an experience or how a recurring idea appears in their responses, qualitative material and coding may fit. Sometimes both are useful. The U.S. Census Bureau’s survey-methods work, for example, uses qualitative and quantitative methods to improve questionnaires, collection instruments, and data quality. The methods are not rival magicians fighting over one rabbit. They can perform different jobs within the same investigation.

Suppose you want to study whether people understand a survey question about “regular household activity.” A numerical result might show how responses are distributed. Qualitative work might help you examine what respondents thought the phrase meant. Did they include taking out the garbage? Feeding a lizard? Standing in the kitchen while somebody else makes eggs? The phrase may look tidy until an actual person gets near it.

That is one reason method selection begins with the question and not with whatever software you already know.

A Survey Is Not Automatically the Answer

Surveys are useful when the evidence you need can reasonably be collected through questions and recorded responses. But “send out a survey” is not a complete research design. It is an activity, like purchasing rope.

The wording and structure of the instrument matter because the instrument is part of the evidence-producing process. The Census Bureau explicitly connects survey-methods research with improving questionnaires, collection instruments, and data quality. In other words, the questions themselves deserve investigation. A confusing item does not become scientific merely because 600 people were confused by it in an orderly fashion.

Before treating survey answers as findings, make the questionnaire reviewable. Preserve the exact wording. Preserve the response choices. Explain how the responses were collected. If wording changed, record what changed instead of quietly replacing the old version and pretending the survey emerged fully formed from an egg.

This documentation also makes later interpretation less mystical. A reviewer should not have to guess whether “often” was an original response category, a later grouping, or something Dennis typed into the chart because he liked the sound of it. There is always a Dennis. Sometimes Dennis is you.

Qualitative Coding Needs a Visible Trail

Qualitative research can involve examining text for ideas, categories, or recurring patterns. The actual material might be open-ended survey responses or some other qualitative data available to the project. The essential practical problem is that the path from text to interpretation must be inspectable.

That means documenting the coding process and preserving the research artifacts needed to understand it. What material was examined? What labels were applied? What did those labels mean? When a passage could fit more than one label, how was that handled?

You do not need to pretend that interpretation disappeared. It did not climb out a bathroom window during lunch. Explain it.

A useful coding record can show the relevant source material, the code applied to it, and any note needed to understand the decision. The purpose is not to produce a giant ceremonial binder that nobody opens. The purpose is to let another person examine how the interpretation was made.

This matters in practical research work. A SaidGig Social Science Research Assistant listing calls for qualitative coding alongside survey analysis, literature reviews, statistical analysis, documented research artifacts, and reproducible work. That single listing does not establish some grand labor-market prophecy. It does show one concrete setting where producing evidence and documenting how it was produced belong to the same job.

For the research-oriented AI evaluation tasks in that cited role, method selection, evidence synthesis, transparent documentation, and defensible interpretation are transferable skills. The listing establishes their relevance to that role only; it does not support a broader labor-market generalization.

Sometimes the Data Already Exists

You may not need to collect new information at all.

Secondary analysis means reusing previously collected qualitative or quantitative data to address a new research question. This can be a sensible match when an existing dataset contains material relevant to what you want to investigate.

But the new question and the old data must actually meet. They cannot merely wave to one another from opposite buses.

Ask what the existing material contains, how it was originally collected, and whether those contents can address your present question. If the dataset contains answers about commuting time, it will not necessarily answer your question about why people enjoy working from home, even if both topics involve pants.

Document which data you used and which parts you analyzed. If you transformed variables, grouped responses, excluded material, or created a smaller analytic set, record those decisions. A reviewer should be able to distinguish the original material from the version you constructed for your analysis.

Literature Review Is Research Work, Not Decorative Reading

A literature review can help you examine what is already known and how earlier work approached the subject. But “I read several things” is not enough of an account.

Keep a record of what you reviewed and how it related to the question. Distinguish an author’s conclusion from your own interpretation of it. If a source was considered but did not answer the question, that can be worth recording too. Otherwise the final review may look like a pile of sources that wandered into the room independently and formed a government.

A literature review may also change the question. Perhaps the original version was too broad. Perhaps existing work already addresses one part, while another part remains useful to investigate. Changing the question is not inherently a failure. Hiding the change is where the little research goblins enter.

Credible Means a Reviewer Can See What Happened

Credibility is not a decorative adjective placed before “research.” It comes from exposing enough of the work for someone else to assess it.

Preserve the question you began with. Identify the evidence used. Keep the questionnaire, coding definitions, analysis materials, and relevant source records. Document changes and decisions. If you conducted statistical analysis, preserve the artifacts needed to understand what was analyzed and how the result was produced. If you interpreted qualitative material, show the connection between the material and the interpretation.

Also write down limitations without acting like you are confessing to robbing a lighthouse. A limitation tells the reviewer what the work can and cannot support. It helps prevent a specific result from inflating itself into a universal announcement.

Then we arrive at three words that are often placed near one another until everybody assumes they are cousins: reproducibility, replicability, and generalizability.

They are different.

The National Science Foundation distinguishes them this way:

  • Reproducibility concerns obtaining the result using the same materials and procedures.
  • Replicability concerns using new data under the same procedures.
  • Generalizability concerns whether findings apply to other contexts or populations.

Satisfying one does not automatically satisfy the others. A result can be reproducible without demonstrating that it applies everywhere. New data can be examined under the same procedures without proving that every other population will behave identically. These are separate questions, and researchers should stop making them share one toothbrush.

Decide which claim you are making. Then provide the material needed to assess that claim.

A Practical Way to Choose

Start with the narrowest honest version of your question.

If you need structured responses suitable for numerical examination, consider survey analysis or other quantitative material. If you need to interpret how people describe something, consider qualitative evidence and documented coding. If relevant data has already been collected, examine whether secondary analysis can answer the new question. If you need to understand existing work, conduct a review that records what you examined and how you used it. If the question genuinely needs more than one kind of evidence, combine approaches and explain what each one contributed.

Then preserve the trail: the question, the materials, the procedures, the transformations, the interpretive decisions, and the limitations.

That trail is not administrative debris left behind after the exciting research gallops away on a white horse. It is part of the research. Without it, a reviewer sees the final result standing alone in a parking lot at night, claiming it came from science.

Maybe it did. Show them how.