Accounting candidates usually choose an analysis method first and then discover the data is unobtainable, unaffordable or locked behind a subscription their faculty does not hold. Reverse that. The table below compares the five approaches that actually appear in South African accounting mini-dissertations, judged on what data each one demands, what that data costs you in rand, and how long it takes to finish.
The comparison at a glance
| Approach | Data required | Where South African students get it | Realistic cost | Time to complete analysis | Difficulty |
|---|---|---|---|---|---|
| 1. Disclosure index / content analysis | Annual and integrated reports for a defined sample of listed or state-owned entities | Company investor-relations pages and the JSE’s published announcements — all free | R0, plus data for downloading large PDFs | Long collection, short analysis | Low statistically, high in labour |
| 2. Cross-sectional OLS regression | One year of financial variables across many firms | A financial database if your faculty subscribes; otherwise hand-collected from reports | R0 if licensed institutionally; unaffordable if not | Fast once data is clean | Moderate |
| 3. Panel regression (fixed or random effects) | Multiple firms across multiple years | Same as above, but multiplied by the number of years | R0 if licensed; weeks of hand-collection if not | Moderate | High |
| 4. Event study | Daily share prices, a market index, and precisely dated announcements | Price data from a subscribed terminal or database; announcement dates from public JSE news services | Effectively subscription-dependent | Fast once the event window is built | High |
| 5. Practitioner survey | Responses from accountants, auditors or finance officers | You collect it yourself, with ethics clearance | R0 to low, using institutional survey software | Slow, and hostage to response rate | Low to moderate |
One rule governs the whole table: check what your faculty already licenses before you design anything. Most South African commerce faculties provide postgraduate access to at least one financial database, and several run terminal rooms where access is free to registered students but only on campus. That single fact decides whether options 2, 3 and 4 are open to you.
The ranked shortlist
1. Disclosure index content analysis — the one that always finishes
Who it suits: anyone without database access, anyone researching integrated reporting, sustainability disclosure, governance code application or the quality of a specific note in the financial statements.
You build a checklist of items from a framework, apply it to each company’s report, score it, and then analyse the scores. Everything you need is free: integrated reports sit on corporate websites, and the governance and reporting frameworks that generate the checklist are publicly documented. Analysis is often descriptive — means, distributions, comparisons across sectors or years — which is entirely sufficient for a research report at honours or coursework master’s level.
Where it falls short: the labour is brutal and front-loaded. Scoring 60 integrated reports against a 40-item index is 2 400 judgements, and every one must be consistent. Coding reliability becomes a real methodological requirement: score a subsample twice, or have a second coder do it, and report the agreement. Examiners ask.
2. Cross-sectional OLS regression — the default, and usually the right one
Who it suits: a candidate with database access and one clear relationship to test, such as whether a firm characteristic is associated with a reporting or performance outcome in a single year.
It is well understood, every supervisor can examine it, and the diagnostics are teachable in an afternoon. If you are still deciding what your question needs statistically, the broader decision logic in the guide to choosing the right statistical test maps cleanly onto accounting variables.
Where it falls short: a single year is vulnerable to whatever happened that year, and accounting variables misbehave. Expect skewed distributions and outliers driven by a handful of very large listed firms. Winsorising and log transformations are standard practice here, and you must report exactly what you did. If your residuals fail their assumptions, do not panic and do not switch tests blindly — the routes forward are laid out in the guide on what to do when your data fails the normality test.
3. Panel regression — powerful, and the most common overreach
Who it suits: doctoral candidates, and master’s candidates with confirmed multi-year database access and a supervisor who has run panel models before.
A panel controls for unobserved firm characteristics that a cross-section cannot, which is a genuine methodological advantage. The cost is a chain of decisions you must defend: fixed versus random effects and the test you used to choose, how you handled entries and exits from the sample, standard error treatment, and what you did about persistence in the dependent variable.
Where it falls short: it is where mini-dissertations go to die. The South African listed universe is small — a few hundred companies on the main board, thinning further once you exclude financials and impose a data-availability filter — and a short, unbalanced panel with a modest firm count delivers less than students expect. Software matters too: this is the point at which SPSS becomes awkward and Stata, EViews or R become the practical choice. The licensing picture for each is set out in the comparison of SPSS, R, jamovi and JASP for South African postgraduates.
4. Event study — high impact, narrow window of feasibility
Who it suits: candidates researching market reaction to announcements — results releases, auditor changes, governance failures, restatements — who have daily price data and confident dating.
Done properly it is the most persuasive design on this list, because it isolates a reaction to a specific moment. The mechanics are standardised: estimation window, event window, expected returns from a market model, abnormal returns, cumulative abnormal returns, significance testing.
Where it falls short: two South African-specific problems. First, thin trading. Small and mid-cap counters on the local market do not trade every day, and non-synchronous trading distorts abnormal return estimates — you will need an explicit liquidity filter and you must justify it. Second, dating. Your result is only as good as your announcement dates, and information often reaches the market before the formal release. Build a clean, auditable event date list before you commit to the design.
5. Practitioner survey — the right answer for behavioural and public-sector topics
Who it suits: topics with no archival footprint — audit judgement, ethical pressure on finance staff, management accounting practice in small enterprises, skills gaps in municipal finance offices.
Archival data cannot tell you why a preparer chose a treatment; a survey or interview can. This route also removes the database dependency entirely.
Where it falls short: the response rate. Chartered accountants and municipal finance officials are difficult to reach and slower to respond than any sample size calculation assumes, and you carry a full ethics clearance process before you may send a single questionnaire. Budget three months from clearance to a usable dataset and start the ethics application early.
The recommendation
For a South African honours research report or coursework master’s mini-dissertation completed in a single academic year, build a disclosure index and analyse it descriptively, or run a cross-sectional OLS regression if your faculty licenses a financial database. Those two carry the lowest risk of a design that cannot be executed, and both produce a defensible research report.
Choose panel regression or an event study only when three things are true at once: your database access is confirmed rather than assumed, your supervisor has examined that design before, and you have a full year for analysis. The most expensive mistake in accounting research is not choosing the simpler method — it is spending four months discovering that the ambitious one was never feasible.
Getting the data without a subscription
If your faculty does not license a financial database, the free layer is larger than most candidates realise. Company integrated reports and annual financial statements are published on corporate websites. Listed-company announcements are distributed through public news services. The central bank and the national statistics agency publish macroeconomic series at no cost, and audit outcome reports for the public sector are published in full each year — the richest free dataset available to anyone researching public-sector accounting. The access procedures for the national statistical sources are covered in the walkthrough on getting South African data for your dissertation.
Two warnings. Hand-collected data must come with a documented collection protocol, because an examiner cannot verify what they cannot reconstruct. And whichever route you take, your methodology chapter has to justify the sample, the period and the exclusions before it justifies the model — the structure for doing that is set out in the guide to building a methodology chapter step by step, which transfers directly to a commerce faculty study.
Write the analysis chapter while the decisions are fresh
Method choices made in March are impossible to reconstruct in September, and examiners fail candidates on undocumented decisions far more often than on the wrong model. Tesify keeps the reasoning attached to the writing: your sample filters, your exclusions, your transformations and your diagnostic results, drafted into the chapter in the order an examiner reads them, with references formatted to your faculty guide. Start your accounting research report with Tesify and get the method section written before the data goes cold.
Frequently asked questions
How many companies do I need for an accounting regression?
It depends on the number of predictors in your model rather than on a fixed rule. Derive the figure from a power analysis for your specific model and then check that the local listed universe can supply it after your exclusions. Many South African studies end with 60 to 150 firms in a single year, which is workable for a small model and restrictive for a large one.
Do I need ethics clearance for archival accounting research?
You still need a formal decision from your faculty, even where the work uses only published data. Most South African faculties process such studies as an expedited or low-risk review and issue a clearance certificate confirming no human participants are involved. You cannot skip the application — the certificate number is required at submission.
Which software should I use for a panel regression?
Stata and EViews are the conventional choices in South African commerce faculties, and R handles panels well at no cost. SPSS can be pushed into it but is not the natural tool. Ask your supervisor which software they can actually examine, because a model no one in the department can check is a liability regardless of its technical merit.
Is a descriptive-only accounting study good enough for a master’s?
For a coursework master’s mini-dissertation, yes, provided the descriptive work is systematic and the interpretation is substantive. A carefully constructed disclosure index applied consistently to a defined sample is a contribution. What fails is descriptive statistics presented without a framework, a rationale or comparison across any meaningful dimension.
How do I handle outliers in financial data?
Winsorising continuous variables at the 1st and 99th percentiles is the convention in accounting research, and log transformation is standard for size measures. Whatever you do, state it in the methodology, apply it consistently, and report results before and after if the treatment materially changes your conclusion. Deleting inconvenient firms without disclosure is a serious integrity problem.
Can I research a private or unlisted South African company?
Only with the company’s written permission, and then usually as a case study rather than a statistical analysis. Unlisted financial statements are not freely available and per-document retrieval from the companies register is slow and costly at scale. A single-firm case study is a legitimate design, but it must be framed as one from the start rather than presented as a sample.
What accounting data is genuinely free in South Africa?
Integrated and annual reports on company websites, listed-company announcements through public news services, central bank and national statistics releases, public-sector audit outcome reports, and municipal and departmental annual reports. That layer is enough to support a full disclosure-based study without any subscription at all.
How far back should my study period go?
Let a reporting change define the boundary rather than choosing a round number. Anchoring the period to the adoption of a standard, a governance code revision or a regulatory change gives you a justified start date and a built-in argument. Three to five years is typical for a master’s study; a single year is acceptable for a cross-sectional design if you justify why that year is representative.
Do I need to test for multicollinearity?
Yes, and report it. Accounting control variables are frequently correlated with one another, so present a correlation matrix and variance inflation factors. Values above 10 are conventionally treated as a problem, and above 5 warrant comment. Deal with it by dropping or combining variables and explaining the decision, not by ignoring the diagnostic.
Should I use a survey or archival data for an auditing topic?
It depends on whether the phenomenon leaves a trace in public documents. Auditor changes, opinion types, tenure and fees where disclosed are archival questions. Judgement, independence pressure and workload are not — they require a survey or interviews, with the response-rate risk and ethics timeline that implies. Choose by what your question needs, then confirm the data route exists before committing.