Quantitative Analysis Is a Sequence, Not a Button
Students often imagine quantitative analysis as a single moment: feed the data into the software, press a button, and out comes the answer. In practice it is a sequence of decisions, and each one affects the next. Clean the data badly and the tests mislead you. Skip the descriptive statistics and you will not notice the outlier driving your headline result. Choose the wrong test and a correct-looking p-value answers a question nobody asked. This guide walks through the sequence in order, so that by the time you report a result, you can defend every step that produced it.
The running example is a simple one: a survey of first-year nursing students measuring academic confidence before and after a study-skills workshop, with a comparison group that did not attend. Keep it in mind as each step is explained — the same logic applies whether your data are survey scores, reaction times, or sales figures.
Step 1: Prepare and Clean Your Data
Analysis begins before any test, with the unglamorous work of getting the data into shape. First, check for missing data: how much is missing, and is it missing randomly or in a pattern? If the students who dropped out of your study are precisely the ones who struggled most, that pattern matters, and you should say so. Common responses are reporting the missing cases, excluding them with a stated rule, or imputing values — but whichever you choose, state it plainly and consistently.
Next, look for errors and outliers. A confidence score of 240 on a 0–30 scale is a data-entry error, not a remarkable student. A genuine extreme value is different: it may be a real, interesting case, and deleting it without justification is as bad as keeping an error. The rule is simple — correct errors, investigate outliers, and document every change you make. Keep an untouched copy of the raw data and a log of cleaning decisions; examiners love an audit trail, and our dissertation methodology chapter guide explains how to present one.
Finally, code and recode. Reverse-scored questionnaire items need reversing and categories need consistent labels. Do this before you look at any results, not after — recoding decisions made with the answer in view are exactly how bias creeps into honest work.
Step 2: Describe Your Data Before You Test Anything
Descriptive statistics are not a warm-up; they are the first analysis, and skipping them is the most common beginner error. For each variable, report the appropriate summary: means and standard deviations for roughly normal continuous data, medians and interquartile ranges when the data are skewed, frequencies and percentages for categories. Plot the data too — a histogram or boxplot will show skew, outliers, and odd patterns that a table of means hides completely.
In the workshop example, the descriptives might show that the workshop group started with higher confidence than the comparison group before any workshop happened. That single observation changes the whole analysis: a simple post-workshop comparison would now be misleading, and you would need to compare change over time instead. You only discover that by describing first and testing second. Scribbr’s guide to quantitative research explains the descriptive–inferential distinction clearly, with the same kind of worked example.
Step 3: Choose the Right Test
Choosing a test is a matter of matching three things: your research question, your variable types, and your design. The table below covers the tests most undergraduate and master’s dissertations actually need.
Comparing two groups on a continuous outcome: an independent-samples t-test (different people in each group) or a paired-samples t-test (same people measured twice). Workshop group versus comparison group on post-workshop confidence? Independent-samples t-test. The same students before and after? Paired-samples t-test.
Comparing three or more groups: a one-way ANOVA, followed by post-hoc tests to find where the differences lie. An ANOVA that is significant without post-hoc tests tells you that some difference exists somewhere — which is almost useless on its own.
Looking for a relationship between two continuous variables: Pearson’s correlation, or a regression if you want to predict one from the other. Remember that correlation is not causation, however strong the number: confidence and workshop attendance may rise together because motivated students both attend and feel confident.
Working with categories: a chi-square test for association between two categorical variables, such as attendance (yes/no) against pass/fail. If your outcome is categorical, t-tests and ANOVAs do not apply, however tempting they look in the software menu.
When in doubt, state your reasoning in the methodology: “A t-test was chosen because the outcome was continuous, the groups were independent, and the data met the test’s assumptions.” That sentence alone earns marks, because it shows the choice was deliberate rather than default.
Step 4: Check the Assumptions (Yes, Really)
Every statistical test comes with assumptions, and examiners do check whether you addressed them. The common ones are normality (roughly bell-shaped distributions for t-tests and ANOVA), homogeneity of variance (similar spread across groups), independence of observations, and, for regression, linearity and the absence of extreme multicollinearity.
Checking does not mean perfection. Real data are rarely perfectly normal, and most tests tolerate moderate violations, especially with larger samples. What matters is that you checked and reported what you found and what you did about it. If variances were unequal, report the Welch-corrected t-test. If data were skewed, consider a transformation or a non-parametric alternative such as the Mann–Whitney U test, and say why. A paragraph that honestly addresses a violated assumption scores far better than silence, because silence reads as ignorance.
Step 5: Run the Analysis and Report It Properly
Now, finally, the test. Report each result completely: the test used, the test statistic with degrees of freedom, the p-value, the effect size, and the descriptive statistics it rests on. For the workshop example: “Students who attended the workshops scored significantly higher on post-workshop confidence (M = 18.7, SD = 2.9) than those who did not (M = 12.4, SD = 2.1), t(98) = 12.19, p < .001, d = 1.02.” A reader can verify, interpret, and judge that sentence. A sentence that reports only “p < .05, significant” cannot be judged at all.
Two reporting rules matter more than any others. First, report effect sizes and confidence intervals alongside p-values, because significance tells you a result is unlikely to be chance, while the effect size tells you whether it matters. Second, report non-significant results with the same completeness. A hypothesis that failed is a finding, not an embarrassment to hide, and our dissertation discussion chapter guide shows how to interpret such results honestly rather than burying them.
A Worked Example, Start to Finish
Back to the workshop study. Cleaning found 6 incomplete surveys, excluded under a stated rule (more than 20% of items missing), leaving 100 students. Descriptives showed the groups were similar at baseline. An independent-samples t-test on post-workshop scores was significant with a large effect size, and the assumption checks passed. The results chapter therefore reports the descriptives, the test, and the verdict: the hypothesis was supported. The discussion then asks why — was it the skills taught, or the peer contact? That question belongs in the discussion precisely because the numbers alone cannot answer it.
Mistakes That Undermine Quantitative Analysis
Testing before describing. Running tests on data you have never plotted is how outliers and coding errors become headline findings.
Running every test in the menu. Fishing through tests until one is significant, then reporting only that one, is p-hacking, and it invalidates the result. Decide your tests from your research questions, before you see the outcomes, and report the planned tests even when they disappoint.
Confusing significance with importance. With a large sample, a trivial difference can be statistically significant. Always ask, alongside the p-value, whether the effect is big enough to matter in practice.
Ignoring assumptions entirely. A test run on data that violate its assumptions is a number without a warrant. Check, report, and adapt.
Reporting only p-values. Without descriptives and effect sizes, your results cannot be evaluated, replicated, or compared with other studies. They are decoration, not evidence.
A Checklist Before You Write Up
- Missing data and outliers were investigated, handled under a stated rule, and documented.
- Every variable is described (with plots) before any test is reported.
- Each test matches the research question, variable types, and design — and the choice is justified.
- Assumptions were checked and any violations addressed or honestly reported.
- Every result reports the test, statistic, degrees of freedom, p-value, effect size, and descriptives.
- Non-significant results are reported as fully as significant ones.
- No test was chosen after peeking at the results.
Quantitative analysis rewards order more than cleverness. Clean first, describe second, test last, and report everything — and the numbers will stand up when the examiner leans on them. If you would like help checking that your tests match your questions, our research proposal guide is a good place to re-check the alignment before you write, and our dissertation conclusion guide shows how analysed results become a final answer.
If your dataset is sitting unanalysed and the deadline is closer than comfortable, you do not have to work through it alone. The Pen Master pairs you with subject-matched writers experienced in quantitative analysis, from cleaning to reporting. Message us on WhatsApp at +49 1521 3962850 or send your brief through the contact form on thepenmaster.com, and we will take it from there.