Which Test Should I Use? — The Built-in Assistant
When you open the Calculator & MyData module, the first card on the main menu is "Which Test Should I Use?" — a built-in helper for finding the right statistical test when you are not sure which one fits your study. It is the answer to the most common question new users have: "I know my data, but I don't know what test to run."
The button on the dashboard shows:
Which Test Should I Use? "Use the decision tree or AI assistant to find the right statistical test"
Clicking it opens the assistant page, titled:
Which Statistical Test Should I Use? "Choose how you'd like to find the right test for your analysis. Use the guided decision tree for a step-by-step approach, or describe your problem to the AI assistant."
Two ways to find your test
The assistant gives you two different approaches to choose from. You pick the one that suits how you like to work:
| Approach | Best for | How it works |
|---|---|---|
| Decision Tree | Anyone — even with no statistical knowledge | Answer a series of guided questions about your data and goals. The tree narrows down to one recommended test |
| AI Assistant | Anyone comfortable describing their study in their own words | Describe your research question in natural language. The AI analyses your description and recommends up to three appropriate tests |
Both approaches end at the same place: a clear recommendation of which test to use, with a direct link to open it. You can always switch between the two by going Back to selection.
Important — what the AI actually does: The AI Assistant recommends which test to use for your study. It does not run the calculations, interpret your results, or analyse your data. Once you have a recommendation, you open the recommended test and run it with your own data. The statistical computations are performed by the test's validated engine.
The Decision Tree
The Decision Tree is a guided, step-by-step wizard that asks you about 5 questions on average and ends with a single recommended test.
Step 1 — Choose your research field
Before any statistical questions, the tree first asks:
"What is your research field?" "This helps us show relevant examples at each step. You can change it anytime."
You pick one of eight domains. The chosen field controls the examples shown at every question — the tree itself does not change, but each option is illustrated with examples relevant to your domain.
| Domain | Description |
|---|---|
| Basic Science | Laboratory, experimental, preclinical research |
| Internal Medicine | Clinical diagnosis, treatment, patient outcomes |
| Surgery | Surgical procedures, operative outcomes |
| Nursing / Care | Patient care, nursing interventions, care quality |
| Health Services | Health policy, management, service delivery |
| Life Science | Biology, microbiology, genetics, ecology |
| Pharma / Clinical Trials | Drug development, pharmacology, clinical trials |
| General | General-purpose statistical analysis |
A small badge with your chosen field stays visible at the top of the wizard, and clicking it lets you change field at any time without losing your progress.
Step 2 — Answer guided questions
After picking your field, the wizard asks:
"What is the main goal of your analysis?"
There are 13 high-level goals to choose from, covering the full scope of the module:
| # | Goal | Where it leads |
|---|---|---|
| 1 | Compare groups (means, medians, proportions) | t-tests, ANOVA, nonparametric alternatives |
| 2 | Examine relationships or correlations between variables | Correlation, regression, canonical correlation |
| 3 | Predict an outcome based on other variables | Linear, multiple, logistic regression, GLM, GEE |
| 4 | Analyze survival / time-to-event data | Kaplan-Meier, Cox, Competing Risk |
| 5 | Assess diagnostic test performance | Diagnostic Accuracy, ROC Curve |
| 6 | Measure reliability or agreement | Cronbach's Alpha, ICC, Kappa, Bland-Altman, Item Analysis |
| 7 | Reduce dimensions or classify / group data | PCA, EFA, Discriminant Analysis, Cluster Analysis, SEM |
| 8 | Conduct an epidemiological study | Cohort, Case-Control, Cross-Sectional, Hardy-Weinberg |
| 9 | Analyze survey or questionnaire data | Chi-Square, Cronbach's Alpha, Item Analysis, EFA |
| 10 | Evaluate a treatment or intervention effect | Treatment Efficacy, McNemar, Logistic Regression |
| 11 | Monitor or improve process / product quality (SPC, QC) | Control Charts, CUSUM/EWMA, Process Capability, Gage R&R, Pareto, Acceptance Sampling |
| 12 | Summarize / describe my data | Descriptive Statistics, Summary Statistics CI |
| 13 | Prepare, transform, or process my data | Tools: Transformations, Filtering, RCT randomisation, Scientific Calculator |
After your first answer, the tree branches into follow-up questions about data type (continuous / categorical / binary / ordinal / count / time-to-event), number of groups, independence (independent / paired / repeated measures), distribution (normal / non-normal), and similar considerations — roughly 5 questions in total before you reach a recommendation.
Navigation during the wizard
While you are answering questions, three controls are always available:
| Control | What it does |
|---|---|
| Field badge (top right) | Shows your chosen domain — click to change field |
| Back button | Goes back one question (and re-enables that question's options) |
| Start Over button | Resets the wizard completely — back to step 1, field selection |
A progress bar at the top fills as you answer each question, and below the current question you see a breadcrumb showing every question you have answered so far together with your chosen answer — so you always know how you got where you are.
Each answer option also shows a small example sentence in italic below it, tailored to your chosen research field. For example, in the Internal Medicine domain, the answer "Compare two independent groups" might show "Comparing average blood pressure between a drug group and a placebo group".
Step 3 — Reach a recommendation
When the tree reaches the end of a path, the wizard shows the Recommended Test screen. This screen contains:
| Element | What it shows |
|---|---|
| Recommendation type label | Either "Recommended Test" or "Recommended Tool" (for items from the Tools section) |
| Test/tool name | The exact name of the recommended calculator (e.g. "Student's t-test") |
| Description | A short explanation of what the test does and why it fits your situation |
| Key Assumptions | A list of statistical assumptions the test relies on (e.g. "Independent groups", "Normally distributed data") |
| Requirements | What you need in your data to run the test (e.g. "Continuous outcome variable", "At least 30 observations per group") |
| Your Path | A breadcrumb summary showing your chosen field plus every question and answer that led to this recommendation |
| "Go to" button | Opens the recommended test directly, ready for you to enter or upload your data |
If the recommendation goes to a different section than expected (for example, an item in the Tools section), the screen makes this clear with the "Recommended Tool" label so you know what to expect.
Tip — explore the alternatives: If you want to compare what a different answer would have led to, click Back repeatedly to return to any earlier question and try a different option. Your previous path is preserved as a breadcrumb until you click Start Over.
Coverage
The Decision Tree covers all 53 tests and tools in the platform. Whatever the right answer is — from a simple t-test to a complex SEM model or a quality-control chart — the tree can lead you to it.
The AI Assistant
The AI Assistant is a chat-style interface where you describe your study in plain language and get back up to three test recommendations. It is titled:
Test Selection Assistant "I'll help you find the right statistical test"
When you open it, the conversation starts with a single greeting:
"Hello! I can help you choose the right statistical test for your analysis. Please describe your research question and data in a few sentences."
How to use it
Type a description of your study in the text area at the bottom. A few examples of useful descriptions:
- "I have 200 patients randomised to two groups, and I measured their blood pressure at 6 months. The data look normally distributed. Which test should I use?"
- "I want to compare survival times between three treatment groups in a cancer trial, with some patients still alive at the end of follow-up."
- "I have responses to a 20-item questionnaire and I want to check whether the items measure a single construct."
Press Enter or click the send button (the paper-plane icon) to submit. Pressing Shift+Enter inserts a new line if you want a longer description.
While the AI is thinking, you see a loading indicator next to its avatar. The reply typically appears within a few seconds.
What you get back
The AI replies with two parts:
-
A short conversational explanation — usually 2–4 sentences explaining its reasoning. For example: "Since you have two independent groups and a normally distributed continuous outcome, an independent-samples t-test is the most appropriate parametric test. If you are unsure about normality, you could also consider the nonparametric Mann-Whitney U Test."
-
A list of recommendations — a panel titled "Recommended Tests" appears below the explanation, listing up to 3 tests the AI considered most appropriate. For each one, you see:
Element What it shows Test name The exact name of the test in the platform Description A brief reason why this test fits your situation Confidence percentage A match score from 0% to 100% indicating how confident the AI is for your specific situation "Go to test" action Clicking the recommendation opens that test directly
Asking follow-up questions
You are not limited to one question — you can keep the conversation going. If the AI needs more information before recommending, it will ask you clarifying questions first (typically 1 or 2) about your data type, number of groups, independence, or normality. Once it has enough information, it will give the recommendations.
You can also ask follow-up questions after a recommendation. For example:
- "What if my data is not normally distributed?"
- "How many participants do I need per group?"
- "Can I also adjust for age and sex as covariates?"
Controls inside the AI Assistant
| Control | What it does |
|---|---|
| Back arrow (top-left) | Resets the conversation back to the greeting and clears recommendations |
| X (top-right) | Closes the assistant and returns to the Which Test Should I Use? selection screen |
| Send button | Submits your message (also activates on Enter) |
"Need professional help?"
At the bottom of the recommendations panel, you will see a small link:
"Need professional help? Request consultancy"
If your study is complex enough that an automated recommendation does not feel sufficient, this link offers a way to reach the e-picos team for personalised statistical consultancy.
What the AI will and will not do
The AI is built with strict rules about what it can recommend:
| The AI will | The AI will not |
|---|---|
| Recommend only tests that actually exist in the Calculator & MyData module | Suggest tests not available in the platform |
| Use exact test and category names that match the module | Make up tool names or invent categories |
| Ask 1–2 clarifying questions if your description is ambiguous | Guess when key information (data type, number of groups, etc.) is missing |
| Recommend up to 3 tests when several are appropriate (e.g. a parametric option plus its nonparametric alternative) | Recommend more than 3 tests at once |
| Explain its reasoning in plain language | Run the statistical analysis itself |
| Reset the conversation when you click the back arrow | Save your conversations between sessions |
When the AI cannot respond
If the AI service is temporarily unavailable or the network fails, you will see a friendly error message in place of the response:
"Sorry, I encountered an error: [details]. Please try again."
In this case, you have two options: try sending the same message again after a moment, or switch over to the Decision Tree (which works fully offline without any AI service).
Which one should you pick?
Both approaches arrive at the same destination — the right test for your data. Pick whichever fits how you think:
| You should pick the Decision Tree if... | You should pick the AI Assistant if... |
|---|---|
| You prefer answering structured questions over typing free text | You prefer describing your study in your own words |
| You want to see what every alternative would have recommended | You want to ask follow-up questions in conversation |
| You want to be sure your situation is covered with no ambiguity | You have a complex study that does not fit a simple yes/no question |
| You want to work fully offline / without depending on the AI service | You want explanations of why a particular test is recommended |
Either way, the final result is the same: a working test, opened in the module, ready for you to enter or upload your data.