3.4 Writing Data Analysis, Results, and Discussion¶
What you will learn on this page
- How to ask AI to suggest appropriate statistical methods from your research question and design
- Generating analysis code (R/Python) and how to validate it
- Prompts for first-pass interpretation support (with caution)
- How to write effect-size interpretation
- How to decide the boundary between Results and Discussion
- How to structure results by research question
- How to report non-significant results appropriately
- How to write Limitations
- Practical ways to write Conclusions and Implications
How this page fits in
The roles, move patterns, and conventional phrases for Results and Discussion are explained in
2.1 The IMRaD Structure and
2.2 Checking Conventional Expressions with Corpora.
For APA-style statistical reporting and figure/table caption conventions, see
3.3 Writing the Methods Section.
This page focuses on a practical workflow from analysis to writing Results and Discussion in English.
A basic stance for AI-assisted analysis¶
Even in data analysis, the principle is the same as in writing: AI outputs are first-pass support, not final decisions.
Ground rules for AI in analysis
"Suggesting candidate methods, generating code skeletons, and guiding output interpretation are OK. Fabricating results or generating interpretations not supported by the data are not."
Proposing statistical methods¶
You can describe your research design and data characteristics and ask AI to propose appropriate analysis methods.
Prompt: propose candidate methods
Based on the research design and data characteristics below,
propose up to three appropriate statistical analysis methods.
For each method, briefly explain:
- why it fits this data
- key assumptions (e.g., normality, homogeneity of variance)
- points to watch out for
Research design: [describe here]
Dependent variable(s): [variable name(s) and measurement scale]
Independent variable(s): [variable name(s) and levels]
Sample size: [N]
Data characteristics: [distribution, presence/absence of missing data, etc.]
Prompt: check the validity of the chosen method
Generating and validating analysis code¶
AI is highly capable of generating analysis code in R or Python, but generated code always requires validation.
Code generation prompts¶
Write R code to perform the following analysis.
- Data is in CSV format with column names: [list column names]
- [Describe the specific analysis]
- Include a summary table of results and visualization (ggplot2)
- Add comments in English
Load any required packages with library() at the top.
Write Python code to perform the following analysis.
- Data is a pandas DataFrame with column names: [list column names]
- [Describe the specific analysis]
- Include result display and matplotlib visualization
- Add comments in English
Validation checklist¶
AI-generated code may look fluent but can contain the following problems:
- Calling non-existent functions or packages (hallucination)
- Incorrect argument specification (especially due to package version differences)
- Omitted preprocessing (missing value handling, data type conversion, etc.)
- Insufficient assumption checking
Always run and verify generated code
Do not blindly trust AI-generated code. Always run it in your own environment and verify that there are no errors and that the output is reasonable. In particular, function specifications can change between package versions.
Debugging support¶
When you encounter errors, it is effective to send the error message along with the code to AI.
I ran the following R code and got an error.
Please explain the cause and how to fix it.
Code:
[Paste code here]
Error message:
[Paste error message here]
Designing the analysis pipeline¶
When performing multiple analyses, having AI organize the overall analysis design can reduce oversights.
Prompt: design an analysis pipeline
Design an analysis pipeline to answer the following research questions.
Include:
(1) The analysis method needed for each RQ
(2) The order of execution (assumption checks → main analysis → post-hoc tests)
(3) Preprocessing required for each analysis
(4) A list of statistics to report
Present in flowchart format.
Research questions:
[Paste RQs here]
Data overview:
[Variable names, measurement scales, sample size, etc.]
Coding support for open-ended response data¶
AI can also be used as first-pass coding support in qualitative data analysis.
Prompt: generate first-pass coding candidates
Code the following open-ended responses by theme.
- First, read through all the data and propose up to 5 major themes
- Assign themes to each response and quote the supporting passage
- For responses that fall under multiple themes, show all of them
- Mark responses where the judgment is uncertain as [needs review]
[Paste open-ended response data here]
For details on the accuracy and limitations of AI coding, and the principle of two-stage operation, see 3.3 Cautions when using AI for qualitative data analysis.
Interpretation support for analysis results¶
You can have AI read statistical output or graphs and obtain first-pass feedback on interpretation.
Prompt: interpret statistical output
The following is the output from running [name of analysis method] in R.
Please interpret this result for an applied linguistics researcher.
- Meaning of the key statistics
- Evaluation of the effect size magnitude
- What the results can and cannot tell us
- Points to note (sample size, assumption satisfaction, etc.)
Do not add new claims; explain only within the range readable from the output.
[Paste statistical output here]
Prompt: read a graph
Tips for using interpretation support
AI interpretations are merely candidates for "how this could be read." It is the researcher's responsibility to judge which interpretation is appropriate in light of their research hypotheses and research questions. One effective approach is to have AI interpret the results without sharing your hypotheses, then compare its interpretation with your own.
Writing effect-size interpretation¶
For statistical reporting formats (notation such as t(28) = 2.45, p = .021, d = 0.65), see 3.3 Statistical Reporting Formats. Here, we focus on how to interpret reported effect sizes within the text.
Benchmarks for effect-size interpretation (rough guide)¶
| Index | Use | Small | Medium | Large |
|---|---|---|---|---|
| Cohen's d | Difference between two group means | 0.2 | 0.5 | 0.8 |
| η² (eta-squared) | Effect size for ANOVA | .01 | .06 | .14 |
| r | Correlation coefficient | .10 | .30 | .50 |
| Cohen's f | ANOVA (f-value) | 0.10 | 0.25 | 0.40 |
Effect-size benchmarks are only rough guides
Cohen's (1988) benchmarks (small/medium/large) are conventional approximations. Judge whether an effect size is "meaningful" by comparing it against the norms of prior research in your field. Plonsky & Oswald (2014) reported that the median Cohen's d in applied linguistics is approximately 0.7, meaning that applying Cohen's benchmarks directly may lead to underestimation.
Examples of embedding effect sizes in text¶
Weak description (numbers only):
The effect size was d = 0.65.
Strong description (including interpretation):
The effect size was medium (d = 0.65), indicating that the treatment group outperformed the control group by approximately two-thirds of a standard deviation. This is comparable to the effect sizes reported in similar intervention studies in L2 vocabulary research (e.g., Smith, 2021; Lee, 2022).
Prompt: generate an effect-size description
Write a sentence or two describing the following statistical result,
including an interpretation of the effect size.
Follow APA 7th edition format and add a brief evaluation of the effect size magnitude.
Do not include comparison with prior studies (I will add that myself later).
Analysis: [method name]
Results: [test statistic, p-value, effect size]
Writing the Results section¶
Structuring results by research question¶
The Results section is most readable when organized in the same order as the RQs presented in the Introduction.
| When there is one RQ | When there are multiple RQs |
|---|---|
| Report in order of descriptive → inferential statistics | Create subsections (subheadings) for each RQ |
| Report in a natural paragraph flow | Within each subsection, report descriptive → inferential statistics |
Prompt: check Results structure against RQs
Check whether the following Results section is organized in the
same order as the research questions presented in the Introduction.
Check the following:
(1) Are the results for each RQ clearly distinguished?
(2) Does the order of results match the order of the RQs?
(3) Is sufficient statistical evidence provided for each RQ?
(4) Are there results unrelated to any RQ mixed in?
Research questions:
[Paste RQs here]
Results:
[Paste Results here]
Converting bullet points to Results prose¶
Convert the following analysis results into natural English prose
for the Results section of a research paper.
- Write in the past tense
- Report statistics in APA 7th edition format (e.g., t(28) = 2.45, p = .021, d = 0.65)
- Do not change any numerical values
- Restrict to reporting results only; do not include interpretation or discussion
- Reference figures/tables in the format (see Figure 1)
[Paste bullet-point results here]
Reporting non-significant results¶
Non-significant results are important information for the study and should be reported appropriately.
| To avoid | Recommended |
|---|---|
| ~~The difference was not significant, so we will not discuss it further.~~ | No statistically significant difference was found between the two groups, t(48) = 1.23, p = .225, d = 0.18. |
| ~~The results failed to show...~~ | The results did not reveal a significant effect of X on Y. |
| ~~(Not reporting non-significant results at all)~~ | Report non-significant results with test statistics and effect sizes |
Non-significant ≠ no effect
p > .05 does not mean "there is no effect"; it means "the effect could not be detected with this sample size." By reporting effect sizes, you allow readers to judge the practical magnitude of the difference.
Prompt: describe a non-significant result
The following statistical result was not significant.
Write an appropriate description for the Results section of an academic paper.
Requirements:
- Do not use "failed to"
- Include the test statistic, p-value, and effect size
- Do not include interpretation in Results (save for Discussion)
- Write in the past tense
Result: [test statistic, p-value, effect size]
The boundary between Results and Discussion¶
The principle is to report "results" in Results and "interpretation" in Discussion, but this boundary is often blurred for beginning writers.
| What belongs in Results | What belongs in Discussion |
|---|---|
| Objective reporting of statistical results | Interpretation of what the results mean |
| Presenting numbers and statistics | Comparison with prior work |
| Reference to figures/tables with brief comments | Theoretical and practical implications |
| Reporting unexpected results (without interpretation) | Attempting to explain unexpected results |
Prompt: check the Results/Discussion boundary
Check whether the following Results section contains any "interpretation"
that should be moved to the Discussion.
Criteria:
- Factual reporting directly readable from the data → Results (OK)
- Explanations of "why this happened" → Discussion (remove from Results)
- Comparison with prior studies → Discussion (remove from Results)
- Interpretive expressions such as "suggest", "imply", "may be because" → Discussion
If any interpretation is mixed in, quote the relevant passage
and suggest how it should be handled in the Discussion.
[Paste Results here]
When combining Results and Discussion
Some journals use a combined "Results and Discussion" section. Even in this case, it is good practice to write so that the "reporting" and "interpretation" of each result are clearly distinguishable. A common approach is to state the result and immediately follow with its interpretation.
Writing the Discussion section¶
The Discussion requires interpretation of results and linkage to prior work, so it is risky to have AI generate the entire section. A safer approach is to build the skeleton yourself and ask AI to render it in English.
Step 1: Build the skeleton¶
Write 1–2 sentences for each of the following elements, corresponding to the six moves described in 2.2 Discussion Moves:
| Element | Corresponding move |
|---|---|
| Summary of key findings (what was found) | Summary of results |
| Interpretation of results (why it happened) | Interpretation of results |
| Agreement/disagreement with prior work | Comparison with prior research |
| Theoretical and practical implications of the results | Presenting implications |
| Limitations of the study | Stating limitations |
| Suggestions for future research | Future research |
Step 2: Ask AI to render it in English¶
The following is the skeleton of the Discussion for a research paper.
Convert it into natural academic English paragraphs, maintaining the flow.
- Include appropriate hedging expressions (suggest, may, appear to, etc.)
- Avoid asserting definitive causal relationships
- Do not add new claims or facts
- Do not insert specific literature citations (author, year)
[Paste skeleton here]
Step 3: Check for logical leaps¶
Read the following Discussion and check for excessive generalization
or logical leaps that are not supported by the data.
Quote the problematic passage(s), explain why they are problematic,
and suggest a direction for revision in one line.
[Paste Discussion here]
Caution with AI-generated Discussion
AI tends to generate "plausible but ungrounded interpretations." Be especially careful about references to things your data do not directly show, superficial connections to prior work, and assertions of causality.
Before / After example for Discussion¶
Before (weak discussion):
The results showed that the experimental group scored higher. This is because the treatment was effective. Previous studies also found similar results.
Problems: Asserts causality ("This is because"). Prior study citation is vague. The data may show a correlation, not a causal relationship.
After (strong discussion):
The experimental group demonstrated significantly higher scores than the control group (d = 0.65). One possible explanation for this finding is that the treatment provided additional opportunities for contextualized practice, which has been shown to facilitate vocabulary retention (Smith, 2020). This interpretation is consistent with the findings of Jones (2021), who reported a similar advantage for learners receiving meaning-focused instruction. However, it should be noted that the present study employed a quasi-experimental design, and therefore causal claims should be made with caution.
Writing Limitations¶
"Limitations," typically written at the end of the Discussion, is a section that reviewers always scrutinize. The key is to acknowledge limitations frankly while maintaining a balance that does not undermine the value of the study.
Common categories of limitations¶
| Category | Example | Typical expression |
|---|---|---|
| Sample limitations | Restricted to a specific population | The sample was limited to university students, which may limit the generalizability of the findings. |
| Data collection limitations | Self-reported data only | The reliance on self-reported data may have introduced response bias. |
| Research design limitations | Cross-sectional design; no causal inference | The cross-sectional design precludes causal inferences. |
| Measurement limitations | Only a single measure used | Only a single measure of X was used, which may not fully capture the construct. |
| External factors | Uncontrolled variables | The study did not control for the potential influence of Y. |
Principles for writing Limitations¶
- After stating a limitation, describe its degree of impact (how the limitation could affect the results)
- Connect limitations to suggestions for future research ("Therefore, future studies should...")
- Do not completely negate the value of your study ("Despite these limitations, this study contributes to...")
Prompt: draft Limitations
Based on the following research overview, draft a Limitations section.
Consider limitations from the following five perspectives:
(1) Sample representativeness (2) Data collection method (3) Research design
(4) Measurement method (5) Uncontrolled variables
For each limitation:
- State the limitation (1 sentence)
- Explain the potential impact on results (1 sentence)
- Suggest a direction for future research (1 sentence)
Research overview:
[Summary of research design, method, and results here]
Prompt: convert Limitations to English with expression check
Writing Conclusion / Implications¶
As noted in 2.1 Conclusion in IMRaD, the Conclusion may be placed at the end of the Discussion or as a separate section. In either case, the role of the Conclusion is to succinctly state "what can be concluded."
Elements to include in a Conclusion¶
| Element | Content | Note |
|---|---|---|
| Summary of key findings | Answer to the RQs in 1–2 sentences | Not a repetition of Results; state the significance briefly |
| Theoretical implications | How the study contributes to existing theory | Avoid overclaiming |
| Practical implications | How the findings are useful for teaching, practice, etc. | Include specific suggestions |
| Future research directions | Next steps building on this study | Correspond to Limitations |
Do not introduce new arguments in the Conclusion
The Conclusion is a "wrap-up" of the arguments developed in the Discussion. Avoid arguments that appear for the first time here or new literature citations.
Prompt: generate a Conclusion
Based on the following Discussion content, write a Conclusion section in English.
Requirements:
- Approximately 150–250 words
- Order: summary of key findings (1–2 sentences) → implications → future research
- Stay within the scope of what was discussed in the Discussion (do not add new arguments)
- May begin with "In conclusion," or "To summarize,"
- Close by emphasizing the positive contribution of the study
[Paste Discussion summary here]
Checking the overall structure of the Discussion¶
The following prompt checks whether each paragraph of the Discussion fulfills an appropriate role.
Prompt: evaluate the overall Discussion structure
For the following Discussion section, classify each paragraph
according to which of the following roles it fulfills.
Role options:
(a) Summary of results (b) Interpretation of results (c) Comparison with prior research
(d) Theoretical implications (e) Practical implications (f) Limitations (g) Future research
(h) Other (if unclassifiable, state the reason)
Then evaluate:
(1) Are any roles missing?
(2) Is the order of roles natural?
(3) Is there an imbalance toward any particular role (e.g., too much interpretation, too little on implications)?
[Paste Discussion here]
Next, for whole-paper revision and submission preparation, go to:
3.5 Revision Techniques