Accountancy students pick the wrong test more often than most, because BSA sub-problems tend to mix respondent-perception data with financial figures in the same study. A survey of financial literacy behaves like an education instrument. A study of internal control weaknesses against reported losses behaves like an economics paper. Using one test for both is where panels start asking uncomfortable questions.
The comparison table
| Test | Sub-problem it answers | Data it needs | Typical BSA use | Where to run it |
|---|---|---|---|---|
| Frequency and percentage | What is the profile of the respondents? | Nominal or ordinal | Year level, sex, employment status, firm size, industry of the MSME | Excel or SPSS |
| Weighted mean and SD | What is the level or extent of X? | Likert, 4 or 5 point | Level of financial literacy; extent of internal control practices; perceived usefulness of cloud accounting | Excel or SPSS |
| Independent samples t-test | Is there a difference between exactly two groups? | Interval DV, two-group IV, roughly normal | Financial literacy of working vs non-working students; compliance of registered vs unregistered sari-sari stores | SPSS, JASP, Excel Data Analysis |
| Paired samples t-test | Did the same group change between two points? | Two measures on the same respondents | Pre-test and post-test in a tax seminar or bookkeeping intervention | SPSS, JASP |
| One-way ANOVA | Is there a difference across three or more groups? | Interval DV, categorical IV with 3+ levels | Internal control practices across micro, small and medium enterprises; literacy across year levels | SPSS, JASP |
| Mann-Whitney U | Two-group difference when normality fails | Ordinal or non-normal interval | Small-sample firm studies, skewed peso amounts | SPSS, JASP |
| Kruskal-Wallis H | Three-group-plus difference when normality fails | Ordinal or non-normal interval | Compliance scores across barangays or industry types | SPSS, JASP |
| Chi-square test of independence | Are two categorical variables related? | Both variables nominal, counts per cell | BIR registration status against use of formal books; review centre attendance against CPALE outcome | SPSS, Excel |
| Pearson r | Is there a relationship between two interval variables? | Both interval, linear, normal | Financial literacy score against saving behaviour score | SPSS, JASP, Excel |
| Spearman rho | Relationship when data is ordinal or non-normal | Ordinal or ranked | Ranked audit risk ratings against ranked control ratings | SPSS, JASP |
| Simple linear regression | How much does one predictor explain the outcome? | Interval DV and IV | Effect of financial literacy on saving behaviour | SPSS, JASP, Excel |
| Multiple linear regression | Which of several predictors matter, holding the others constant? | Interval DV, two or more IVs | Determinants of tax compliance among BMBEs; drivers of cloud accounting adoption | SPSS, JASP |
How to read that table against your own sub-problems
Three questions settle almost every choice, in this order.
What is the verb in your sub-problem? “Determine the level of” means descriptive statistics. “Determine whether there is a significant difference” means a difference test. “Determine whether there is a significant relationship” means correlation. “Determine which factors significantly influence” means regression. Panels read the verb, so your test has to match it literally.
How many groups are you comparing? Two groups is a t-test. Three or more is ANOVA. Running three separate t-tests instead of one ANOVA inflates your error rate, and an accountancy panel with a statistics background will catch it.
What kind of data did your instrument actually produce? Likert responses treated as interval support means, t-tests, ANOVA and Pearson r. Yes-or-no and category answers support only frequency and chi-square. If you designed your instrument before choosing your tests, expect to discover a mismatch here — the fix is to check the instrument against the sub-problems the way the item-by-item questionnaire build for education studies lays it out, since the logic is identical whatever your course.
The ranked shortlist for BSA studies
1. Weighted mean with standard deviation — the workhorse
Nearly every BSA thesis with a survey component needs this, and many need nothing more sophisticated for their first two or three sub-problems. Report the mean per item, the mean per dimension, the standard deviation, and the verbal interpretation.
Suits: any study measuring level, extent or degree. Falls short: it describes and nothing else. A thesis that stops at weighted means will be asked at the defense why it never tested anything.
2. One-way ANOVA — the most common upgrade
The moment your fourth sub-problem groups respondents by year level, firm size or industry, you need ANOVA rather than a stack of t-tests. Always run a post-hoc test — Tukey when group variances are equal, Games-Howell when they are not — because a significant F only tells you that some pair differs, not which.
Suits: comparing three or more categories on a Likert-derived score. Falls short: it assumes normality and homogeneity of variance, and small unbalanced groups in MSME samples routinely violate both.
3. Multiple linear regression — where BSA theses earn their marks
If your title contains the words “determinants,” “factors affecting” or “influence of,” regression is not optional. It is also the test that separates a competent accountancy thesis from a descriptive one, because it answers the question a practitioner would actually ask: given everything else, what moves compliance?
Report the R², the F statistic for the model, and the standardised beta with its p-value for each predictor. Check multicollinearity with the VIF; values above 10 mean two of your predictors are measuring the same thing.
Suits: studies with one clear outcome variable and several plausible causes. Falls short: it needs a decent sample — a rough working rule is at least 10 to 15 respondents per predictor — and it is easy to over-claim causation from cross-sectional survey data.
4. Chi-square test of independence — underused in accountancy
Chi-square fits BSA questions that others force into the wrong test. Registration status against bookkeeping method. Firm size against whether an external auditor is engaged. Review centre attendance against CPALE result. All of these are category-against-category, and no amount of averaging makes them anything else.
Suits: two nominal variables. Falls short: it needs an expected count of at least five in most cells, which collapses quickly if you have many categories and a modest sample.
5. Mann-Whitney U and Kruskal-Wallis H — the fallback you should plan for
Peso amounts, firm revenues and audit findings are skewed almost by nature. When the Shapiro-Wilk test rejects normality, these are the non-parametric counterparts to the t-test and ANOVA, and using them is a sign of competence rather than a downgrade. The decision rule and the write-up wording are covered in detail in the guide on what to do when thesis data is not normally distributed.
Suits: small samples, ordinal data, skewed financial figures. Falls short: they compare distributions rather than means, so your Chapter 4 sentence has to be phrased accordingly, and they carry less statistical power.
The recommendation
For the typical undergraduate BSA thesis surveying students, MSME owners or accounting practitioners, use frequency and percentage for the profile, weighted mean with standard deviation for every level-or-extent sub-problem, and one-way ANOVA with a post-hoc test for the comparison sub-problem — then run Shapiro-Wilk first and switch to Kruskal-Wallis if normality fails. Add multiple linear regression only if your title genuinely promises determinants or influence, and only if your sample supports it.
That combination covers the overwhelming majority of accountancy sub-problems, is defensible in front of a panel, and runs in either SPSS or free JASP. Resist the temptation to add structural equation modelling to look sophisticated; panels ask harder questions about tests you cannot fully explain.
Writing the statistical treatment section itself
Your Chapter 3 statistical treatment section is not a list of test names. Each entry names the test, states which sub-problem it answers, and gives the decision rule. A usable pattern:
Weighted mean. This was used to determine the extent of internal control practices among the respondent MSMEs, as stated in sub-problem 2. Responses were interpreted using a four-point scale where 3.26–4.00 indicates Always Practised.
One-way analysis of variance. This was used to determine whether a significant difference exists in the extent of internal control practices when respondents are grouped by enterprise size, as stated in sub-problem 4. The null hypothesis was tested at the .05 level of significance, with Tukey’s HSD applied for post-hoc comparison.
Write one such entry per sub-problem, in the order of the sub-problems, with nothing left over. Anything in your treatment section that no sub-problem asks for is a question waiting to be asked at the defense. The full anatomy of that section, including the significance-level conventions Philippine panels expect, is set out in the explainer on what statistical treatment of data means in a thesis.
Which software to use
Excel handles frequency, percentage, weighted mean, standard deviation, Pearson r and simple regression through the Data Analysis ToolPak, and it is free on any school machine. It does not do post-hoc tests, non-parametric tests or reliability analysis cleanly. SPSS does everything a BSA thesis needs and is what most Philippine research offices expect to see cited; JASP is free, open source and produces APA-formatted output tables. The trade-offs are laid out in the head-to-head on SPSS versus Excel for thesis statistics.
Whatever you use, the output is not your Chapter 4. Converting SPSS tables into the numbered, titled, interpreted tables a Philippine panel expects is a separate job, walked through in the guide to writing Chapter 4 table by table.
Getting the section drafted
Choosing the test is judgement. Writing the paragraph that justifies it, in the phrasing your panel expects, with the decision rule and the significance level stated, is formatting — and it is where accountancy groups lose evenings they do not have.
Tesify drafts your statistical treatment section from your own sub-problems, keeps the wording consistent between Chapter 3 and Chapter 4, and formats your references in APA 7th along the way. Start your thesis on Tesify and get the section written tonight.
Frequently asked questions
What is the most common statistical treatment in a BSA thesis?
Frequency and percentage for the respondent profile, weighted mean with standard deviation for level or extent sub-problems, and one-way ANOVA with a post-hoc test for group comparisons. That trio answers the majority of undergraduate accountancy sub-problems and is straightforward to defend.
When do I use ANOVA instead of a t-test?
Use a t-test when you are comparing exactly two groups and one-way ANOVA when you have three or more. Running several t-tests instead of one ANOVA inflates the chance of a false significant result, and panels with a statistics background routinely flag it.
Do I need regression for an accountancy thesis?
Only if your title or sub-problems promise it — words like determinants, factors affecting, influence or predictors commit you to regression. If your study only measures levels and compares groups, adding regression creates questions you will have to answer without any benefit to the findings.
How many respondents does multiple regression need?
A common working rule is 10 to 15 respondents per predictor variable, so a model with four predictors wants roughly 40 to 60 respondents as an absolute floor and comfortably more in practice. Report your sample size alongside the model and acknowledge the limitation if it is tight.
Can I use chi-square on Likert responses?
You can, by treating each response option as a category, but you usually should not. Chi-square discards the ordering information in a Likert scale. Use weighted mean and a difference test instead, and reserve chi-square for genuinely nominal variables such as registration status or industry type.
What level of significance should I use?
The .05 level is standard across Philippine undergraduate theses unless your research manual or adviser specifies otherwise. State it explicitly in your statistical treatment section and apply it consistently, then report the actual p-value in Chapter 4 rather than only writing significant or not significant.
What if my financial data fails the normality test?
Switch to the non-parametric counterpart: Mann-Whitney U in place of the independent t-test, Kruskal-Wallis H in place of one-way ANOVA, and Spearman rho in place of Pearson r. Skewed peso amounts fail normality routinely, so plan for this rather than treating it as a failure.