Collecting data is only half the work. What happens between raw data and your results chapter determines whether your findings hold up under questioning. We clean, code, and analyze both quantitative and qualitative data, and document every step so your process is reproducible.
| Step | Why It Matters |
|---|---|
| Data cleaning | Missing values, outliers, and entry errors can distort results if left unaddressed |
| Variable coding | Categorical and ordinal variables need consistent, documented coding schemes |
| Assumption checking | Confirms the planned statistical test is actually appropriate for this dataset |
| Transcription (qualitative) | Accurate, verbatim transcripts are the foundation for trustworthy coding |
Skipping this stage is the most common reason a results chapter falls apart under questioning. An output table with no record of how the data was prepared looks suspicious to a careful reviewer, even when the underlying numbers are correct.
Data cleaning is rarely glamorous, but it is where most avoidable analysis errors originate. A messy dataset might include duplicate survey submissions from a respondent who started the questionnaire twice, impossible values (a participant's age entered as "999" because a required field auto-filled when left blank), or missing responses scattered unevenly across variables. Before any statistical test runs or any transcript gets coded, someone has to make a series of judgment calls: which cases get excluded, which missing values get imputed versus left as missing, and which outliers are genuine extreme scores versus data-entry mistakes. None of these decisions is inherently right or wrong, but they need to be made consistently and recorded, because a committee member who spots an inconsistency between two tables in your results chapter will often trace it back to an undocumented cleaning decision made weeks earlier.
Variable coding matters just as much in a quantitative dataset as in a qualitative one, though it looks different. A five-point Likert scale needs a documented direction (does 5 mean "strongly agree" or "strongly disagree"?), reverse-scored items need to be flagged and recoded before they're folded into a composite score, and categorical variables like employment status or treatment group need a coding key that stays consistent from the raw dataset all the way through to the final tables. A coding scheme that drifts partway through analysis, even by accident, is one of the fastest ways to introduce silent errors into a results chapter that otherwise looks polished.
Assumption checking is the step students are most tempted to skip, and exactly the step a statistically literate committee member is most likely to ask about. Every common test, t-test, ANOVA, regression, chi-square, rests on assumptions about the data: normality, homogeneity of variance, independence of observations, adequate expected cell counts. Skipping this check doesn't just risk a missing footnote; it risks running a test that produces a p-value that doesn't mean what it claims to mean. We run and report the relevant checks (Levene's test, Shapiro-Wilk, variance inflation factors for multicollinearity, and so on) before the primary analysis, and note explicitly when a non-parametric alternative was substituted because an assumption didn't hold.
Transcription, for qualitative and mixed-methods work, is the equivalent of data cleaning. A transcript that paraphrases instead of transcribing verbatim, or one that quietly omits pauses, filler words, and non-verbal cues the researcher intended to analyze, has already lost information before coding even begins. We transcribe verbatim by default and flag any passage where audio quality left a word or phrase uncertain, so the coding that follows is built on an accurate record rather than a cleaned-up approximation of what was actually said.
Not every dissertation needs the same kind of analysis, and this page is deliberately the umbrella overview rather than a deep dive into any single method. The deeper coverage lives in more specific guides, linked below, so you're not reading the same explanation twice. If your data is numeric, survey scale responses, test scores, institutional records, you're working in the quantitative tradition, and the detailed treatment of test selection, assumption checking, and reporting statistical output without over-claiming lives in our Quantitative Dissertation Help and Statistics Help guides. If your data is interview transcripts, focus-group recordings, open-ended survey responses, or documents, you're in the qualitative tradition, and the step-by-step coding process, open coding, axial coding, codebook development, saturation, lives in our Coding & Thematic Analysis guide. Many doctoral studies, particularly in education, nursing, and business, use a mixed-methods design that combines both strands, either concurrently or sequentially; our Mixed Methods Dissertation Help guide covers how those two strands get integrated rather than simply reported side by side in the same chapter.
Consider a doctoral candidate, call her Maria, completing an EdD dissertation examining the relationship between teacher self-efficacy and years of classroom experience across three school districts. Her raw data arrives as a survey-platform export: 214 rows, one per respondent, spanning demographic items, a validated 24-item self-efficacy scale, and two open-ended questions about classroom challenges. Before any analysis begins, the dataset needs work. Eleven rows are incomplete surveys where the respondent stopped after the demographics section; these get excluded, and the exclusion is logged with a reason so it can be defended later. Three reverse-scored items on the self-efficacy scale get recoded so that a higher raw score always means higher efficacy, matching how the instrument's original authors intended it to be scored. Years of experience, originally an open text field, gets recoded into the ordinal bands her methodology chapter promised, so the analysis matches what was proposed and approved, rather than drifting into a different scheme invented on the fly.
Once the dataset is clean and coded, Maria runs a one-way ANOVA comparing mean self-efficacy across four experience bands, after first checking Levene's test for homogeneity of variance (it holds, so the standard ANOVA is appropriate rather than a Welch correction). The two open-ended questions get pulled into a separate qualitative workflow, familiarization, open coding, theme development, that runs in parallel with the quantitative analysis, since her study is mixed-methods. The end result is a documented cleaning log, a coding key for every recoded variable, an assumption check reported alongside the ANOVA output, and a set of qualitative themes ready to be integrated with the quantitative findings when she writes her results chapter. Nothing in this process is unusual, but each step is exactly the kind of decision a committee will ask about if it isn't already written down somewhere they can see it.
Master's-level data analysis is often judged mainly on whether the numbers or themes are correct. Doctoral-level analysis is judged on whether the process that produced them would survive an independent replication attempt. That difference shows up in several concrete ways: committees expect a documented audit trail, not just a results table; they expect effect sizes and confidence intervals alongside significance tests, not p-values in isolation; and for qualitative work, they expect explicit attention to trustworthiness, credibility, transferability, dependability, and confirmability, rather than an assumed assurance that the themes are trustworthy because the researcher says so. Doctoral committees also expect the data-analysis approach to be defended twice: once at the proposal stage, as a plan, and again at the defense, as an executed process that matched the plan or explained, with justification, where and why it deviated.
Data analysis is not the finish line. It's the raw material for the chapter that follows. For quantitative and mixed-methods studies, the statistical output gets translated into the objective, question-by-question reporting covered in our Results Chapter guide. For qualitative studies, the coded themes get translated into the narrative, quote-supported presentation covered in our Results/Findings Chapter guide, which explains how "findings" framing differs from "results" framing. Whichever path applies to your study, the handoff works best when the analysis stage already anticipated what the next chapter needs, which is why we treat analysis and results-chapter drafting as one continuous conversation rather than two disconnected deliverables.
Every decision made during cleaning, coding, and analysis should leave a paper trail a stranger could follow. This isn't bureaucracy for its own sake. It's what lets you answer a pointed committee question in real time during your defense instead of promising to "check and get back to them."
Your methods chapter should preview what your analysis will do; your results chapter should match it exactly. A common red flag for committees is a results chapter using an analysis approach that wasn't described, or was described differently, in the methodology. We check this alignment before delivery.
Cleaning, coding, analysis, and a clear audit trail. Ready for committee scrutiny.
Yes. Most of our data analysis work starts from a completed dataset, whether that's survey responses, interview transcripts, or secondary/archival data.
Yes, alongside the written results, so you have both the narrative explanation and the underlying files your committee may ask to see.
Non-significant or unexpected results are reportable findings, not failures. We write the results and discussion honestly, framing what the data actually shows rather than overstating support that isn't there.
It depends heavily on dataset size and method, but most quantitative datasets take a few days once cleaning and coding are done, while a full qualitative coding pass on 15-20 interview transcripts typically takes one to two weeks to do properly. Rushing this stage is where quality slips.
Yes. This decision should already be anchored in your methodology chapter, but when it isn't, or when the proposed approach doesn't quite fit the data you actually collected, we help you choose (and document the justification for) the approach that fits.