系统综述应该有几个研究问题？SciSpace 网络研讨会给出 2–6 个的建议

In short: Usually two to six. There is no fixed rule, but beyond six a single systematic review becomes too big to answer each question in depth. Fewer, deeper questions beat many shallow ones: agree them with your supervisor, then pilot them on 8–10 papers you already know before you build the search.

This answer comes from Dr. Faheem Ullah (University of Adelaide), who covered it in a SciSpace webinar on writing a systematic literature review in 8 steps.

What the expert says

Watch the answer: 9:56 – how many research questions a systematic review should have

Dr. Ullah has read systematic reviews with as few as two research questions and as many as seven or eight. His advice is to stay between two and six and to answer each one in depth, because surface-level answers rarely lead to concrete findings.

“If you define the wrong questions, you end up spending three or four months on something which is not going to be of any use.” — Dr. Faheem Ullah

That is why he puts the questions before everything else. A review done single-handedly takes at least three to four months, so the questions should be agreed with a supervisor or peers before any searching starts.

How to decide how many you need

Start from the purpose of the review, then check each question against the time and people you have.

These ranges are a planning guide based on the webinar advice, not a reporting rule. PRISMA 2020 asks you to state the review's objectives clearly; it does not set a number of questions.

Examples

A real two-question review

Dr. Ullah's systematic review from his PhD, on big data cybersecurity analytics (BDCA) systems, asked just two questions:

1. What are the most important quality attributes for BDCA systems?
2. What are the architectural tactics for addressing those quality concerns?

The second question builds on the first, so the review stays focused and each answer can go deep.

Too broad, then fixed

Before: How is AI used in education? One question, but far too broad to answer in depth.

After: three focused questions that each need different data:

• RQ1: Which AI feedback tools have been tested in undergraduate writing courses?
• RQ2: What effects on writing scores do these studies report?
• RQ3: Which study designs and samples were used?

How to write your research questions

1. Write one main question first. It should match the purpose of the review: a gap, a field overview or a thesis chapter.
2. Add sub-questions only if each needs different data. For example effects, study designs or remaining gaps.
3. Use a framework. PICO for interventions, PECO for exposures and SPIDER for qualitative evidence. See PICO question examples.
4. Check the literature can answer them. Pilot the questions on 8–10 papers you already know are relevant.
5. Get them approved. Your supervisor or peers should agree them before you design the search string.

Common mistakes

• Saturated questions that recent reviews have already answered. Search for existing reviews first, and check whether your angle is new using research gap examples.
• Questions with no literature to answer them. The pilot step catches these early.
• Too many questions, which leads to shallow answers and a review that runs far past its deadline.
• Changing questions after screening starts, which breaks the method and has to be reported as a deviation from the protocol.

Agents to use for this step

• Research Question Generator: generate, refine and narrow research questions for your topic.
• PICO Generator: break each question into population, intervention, comparison and outcome.
• Systematic Literature Review agent: take your final questions through search, screening, extraction and PRISMA reporting.

Related guides

Frequently asked questions

注意转载授权 · 阅读原文