Looking for step-by-step instructions? The full User Guide covers every module in detail.

Open User Guide →
Getting Started
What is AnalyZ Solutions?
AnalyZ is a browser-based data analysis platform that lets you work with both quantitative and qualitative data, no coding or software installation required. It covers the full research workflow: data cleaning, statistical analysis, visualisation, qualitative coding, and sample size planning.
Is AnalyZ free?
Yes, all core analysis features are completely free for registered users. No credit card required. The AI features have monthly caps on the free plan: 1 Sense Making interpretation (refining one also counts), 1 dashboard interpretation, 2 Canvas story generations and 5 Canvas slide refinements. Paid plans with higher limits are available (see Pricing & Plans). A Guest mode is also available with limited features (50 rows / 1 MB) if you want to explore without registering.
What file formats does AnalyZ support?
For quantitative data: CSV, Excel (.xlsx / .xls), Stata (.dta), SPSS (.sav), and JSON. For qualitative data: plain text (.txt), Word documents (.docx), PDF (.pdf) and CSV (.csv). Variable labels and value labels are imported automatically where available.

You can also connect live to survey platforms via API, including KoBo Toolbox, ODK Central, REDCap, Google Sheets, and any custom REST endpoint returning JSON or CSV. Labels, value codes, and multilingual question text are all pulled automatically.
Do I need to know statistics to use AnalyZ?
No. Every module includes a plain-English interpretation panel explaining what the results mean and what you should do next. The Guided Analysis feature also recommends the right test for your goal if you are unsure where to start.
Is my data secure?
Yes. Your data is processed entirely within your browser session and is never stored on our servers. All communication is encrypted with TLS (HTTPS). By default, when you close the tab, the data is cleared from memory. If you'd like to resume a session later on the same device, you can turn on Save on this device in the sidebar, this keeps a copy in your browser's local storage only (never sent to us) until you turn it off or it expires after 24 hours. Saved .analyz.json session files are downloaded directly to your device, no server copy is kept. See the User Guide (Account & Security section) for full details.
Does AnalyZ remember my session if I close the browser?
Only if you turn it on. By default, nothing survives closing the tab. If you'd rather not re-upload every time, toggle Save on this device in the sidebar, the first time you have data loaded, AnalyZ will ask whether you want to keep it locally. This data is stored only in your browser (never sent to our servers), automatically expires after 24 hours, and you can turn it off anytime from the same toggle, which also clears anything already saved.
How do I create an account?
Click Sign up free on the landing page or login screen. Enter your email and choose a password. A verification link will be sent to your email, click it to activate your account. You can then log in at analyz.solutions.
I forgot my password. How do I reset it?
Click Forgot password? on the login screen and enter your email address. A password reset link will be sent to you. The link is single-use and expires after one hour. If it expires, request a new one from the same screen.
Can I use AnalyZ for my thesis or journal submission?
Absolutely. Results tables are formatted to APA standards where applicable. Download tables as Excel or Word files ready for your manuscript. Charts export as high-resolution PNG or interactive HTML.
Can I connect directly to KoBo, ODK, or other survey platforms?
Yes. AnalyZ has a built-in API connector that lets you pull data directly from survey platforms, including KoBo Toolbox, ODK Central, REDCap, and Google Sheets, without downloading any files first.

For KoBo-based platforms, you can sign in with your API token (one click from your KoBo account settings) or with your KoBo username and password: AnalyZ exchanges these for a token automatically, so you never need to hunt one down manually. AnalyZ automatically imports variable labels, value codes, and multilingual question text. You can select which variables to load before the data enters the workspace.

For any other platform, use the Custom API option and paste any REST endpoint URL returning JSON or CSV.
How do I save and restore my session?
Use the 💾 Save Session button in the sidebar after loading data and performing at least one action, this downloads a .analyz.json file. This saves your dataset, variable labels, and action history. Click 📂 Restore Session anytime and select that file to resume exactly where you left off, it works regardless of whether you're currently in Quantitative or Qualitative mode. For qualitative sessions, the saved file also includes your codebook and all coded segments.
Can I filter my data before running an analysis?
Yes. Every quantitative analysis module has a 🔍 Filter data panel at the top. Add one or more conditions (e.g. gender == Female, age > 30) to restrict the analysis to a specific subset without permanently changing the dataset.
Does AnalyZ work on mobile?
AnalyZ is optimised for desktop use. It will load on mobile browsers but the analysis interface, particularly the qualitative coding workspace, is best experienced on a larger screen.
What is Guest mode?
Guest mode lets you explore AnalyZ without an account. It is limited to 50 rows / 1 MB for quantitative data and 1 file / 1 MB for qualitative data. Session saving is disabled. Sign up free to remove all limits.
Data Management
How do I recode a variable?
Go to Data Management → Recode Variable. Select the source variable, define old-value-to-new-value mappings, give the recoded variable a name, and click Apply. The original variable is preserved and the new version is added as a separate column.
How do I create a new variable from existing ones?
Go to Data Management → Create Variable. Choose a method:
  • Formula: arithmetic using existing columns (e.g. income / age)
  • Constant: a fixed value for every row
  • Conditional: if/else logic (e.g. score > 50 → "Pass")
  • Sum / Mean of items: combine scale items into a composite score
Can I merge two datasets?
Yes. Data Management → Merge Data lets you join two datasets on a common key variable. Choose inner, left, right, or outer joins. Append Data stacks two datasets with the same columns on top of each other, useful for combining survey waves.
How do I split a column that contains multiple values in one cell?
Go to Data Management → Split Column. Select the text column, choose a delimiter (comma, space, semicolon, pipe, tab, or custom), and click Preview split to verify before applying. Each part becomes a new column named colname_A, colname_B, etc., inserted right after the source column.
How do I reshape data between wide and long formats?
Go to Data Management → Reshape Data. Choose a direction:
  • Wide → Long: select ID columns to keep and value columns to unpivot (uses pd.melt). Useful for repeated-measures or longitudinal data.
  • Long → Wide: select an index column, a category column (becomes column headers), and a values column (uses pd.pivot_table). Useful for creating summary matrices.
Variable and value labels are cleared after reshaping, re-add them in Data View if needed.
How do I undo a data management action?
Use the ↩ Undo button in the sidebar. AnalyZ keeps a history of up to 20 actions, and ↪ Redo reapplies an undone action. Each undo step restores the dataset, variable labels, and value labels together.
How do I add value labels to numeric codes?
Go to Data Management → Manage Labels. Use the Template tab to download a CSV listing every numeric-coded column and its values, fill in the labels, and upload it back (value labels are replaced for each column in the file). Or use the Manual editor to type labels for one column at a time. Text columns are not listed; convert them first with Encode Text Variable.
How do I turn a text column like Male/Female into numeric codes?
Go to Data Management → Encode Text Variable, select the column and apply. Codes are assigned alphabetically (1, 2, 3 and so on) and the original text becomes the value labels. Review the mapping first, because codes cannot be reordered afterwards.
Can I sort my data or back up a column before changing it?
Yes. Sort Data in Data Management sorts by one or more variables, ascending or descending, with blanks always last. Duplicate Column copies a column (with its labels and type) next to the original as name_copy. Use Undo to revert either.
What can I do by right-clicking a column in Data View?
Right-click a column header for quick actions: change variable type, edit labels, rename, recode, delete, duplicate, split or encode (the last two for text columns), sort ascending or descending, hide the column, or open its frequency table. Each header also has an Excel-style value filter. Filters and hidden columns only affect the table view and never change your data.
How do I set whether a variable is categorical or continuous?
There are two ways. From Data View: click a column header to open the Column Inspector, then click Change in the Variable type row. From Data Management → Variable Types: select a variable and choose its type from the list. Both routes update the same setting.

AnalyZ auto-detects types, numeric columns with more than 10 unique values are treated as continuous, others as categorical, but you can override any variable. This matters for variables like Likert scales coded 1–5, which should be treated as categorical for most tests.

Type overrides apply across all analytical modules (Descriptive Analysis, Inferential, Visualization) for the rest of the session. They survive a tab refresh if Save on this device is on, and are included in any .analyz.json session file you download.
How do I download my cleaned dataset?
Expand the ⬇ Download Data section in the sidebar. Choose a format, CSV, Excel (.xlsx), Stata (.dta), or SPSS (.sav), give it a filename, and click Download. Variable labels and value labels are embedded in the Stata and SPSS exports.
Descriptive Analysis
What descriptive statistics does AnalyZ produce?
Descriptive Analysis has five analysis types: Frequency tables (counts, %, cumulative %), Cross-Tabulate (contingency tables with chi-square), Table (the academic "Table 1", with means, counts and optional p-values by group), Summary Statistics (N, missing, mean, median, mode, SD, variance, min, max, range, skewness, kurtosis, optionally by group), and Data Quality (missing values, duplicates, constants, and outliers for every variable).
How do I check data quality?
Go to Descriptive Analysis → Data Quality. It has four tabs. Overview audits every variable for missing values (count and %), duplicate values, constant columns and outliers beyond 3 standard deviations, colour-coded red, orange or green. Duplicate Rows finds identical rows. Value Issues checks numeric values against range rules you set (or auto-detected implausible values) and tests for digit preference. Enumerator Patterns compares enumerators on straight-lining and interview duration once you pick an Enumerator ID variable in Configure. Run this before any analysis.
How do I run a crosstab and what does it tell me?
Select Cross-Tabulate, choose a row variable and a column variable (both categorical), and click Generate. The output is a contingency table showing counts and row, column or total percentages. A chi-square test of independence is included automatically, it tells you whether the two variables are statistically associated.
Inferential Analysis
What statistical tests are available?
AnalyZ includes 22 tests across five groups:
  • Compare Means: Independent t-test, One-sample t-test, Paired t-test, One-way ANOVA, Two-way ANOVA, Repeated-Measures ANOVA
  • Compare Percentages: Chi-square test, Z-test for Proportion
  • Correlation: Pearson, Spearman
  • Regression: Linear, Logistic, Probit, Multinomial, Ordinal, Hierarchical, Multilevel, Poisson, Negative Binomial
  • Non-parametric Tests: Mann-Whitney U, Wilcoxon Signed-Rank, Kruskal-Wallis, Friedman Test
All tests include plain-English interpretation, effect sizes, and confidence intervals where applicable.
How do I know which test to use?
Use 🧭 Guided Analysis in the top navigation bar, it asks two questions and recommends the right test. As a quick guide: comparing two group means → Independent t-test; three or more groups → One-way ANOVA; two categorical variables → Chi-square; predicting a numeric outcome → Linear Regression; predicting a yes/no outcome → Logistic Regression; count outcomes → Poisson or Negative Binomial Regression; non-normal data → non-parametric alternatives (Mann-Whitney U, Kruskal-Wallis, Wilcoxon, Friedman).
Does AnalyZ report effect sizes?
Yes. Every inferential test reports the appropriate effect size, Cohen's d for t-tests, eta-squared (η²) for ANOVA and Repeated-Measures ANOVA, Cramér's V for chi-square, R² for regression, rank-biserial r for Mann-Whitney U and Wilcoxon, eta-squared for Kruskal-Wallis, and Kendall's W for the Friedman test, along with a plain-English interpretation of whether the effect is small, medium, or large.
What does "ns" mean in the results?
"ns" stands for "not significant" (p ≥ 0.05). AnalyZ uses standard significance labels: *** (p < .001), ** (p < .01), * (p < .05), ns (p ≥ .05). Statistical significance does not equal practical importance, always report effect sizes alongside p-values.
How does Hierarchical Regression work?
Hierarchical Regression lets you build a model in blocks, testing how much each additional set of predictors improves R² (ΔR²). Add control variables in Block 1, your variables of interest in Block 2 (and more blocks if needed), then click Run. The output shows each block's model and a change statistics table showing ΔR² and ΔF for each step.
What is Multilevel Regression and when should I use it?
Multilevel Regression, also called mixed models or random effects models, is used when your observations are nested within groups: for example, patients within clinics, students within schools, or survey respondents within villages. Standard regression assumes all observations are independent; when they are clustered, ignoring the grouping inflates Type I error. Multilevel regression models this clustering explicitly.

Use it when your data has a hierarchical structure, when you want to know how much of the outcome variance is between groups versus within groups (the ICC), or when you have both individual-level and group-level predictors.
What outcome types does Multilevel Regression support?
AnalyZ supports three outcome types in the Multilevel Regression module:
  • Continuous: a numeric outcome such as a score, income, or expenditure. Uses EM-REML estimation.
  • Binary: a yes/no or 0/1 outcome. Uses Penalised Quasi-Likelihood (PQL) and reports odds ratios (OR) or log-odds coefficients, you toggle between them in the results.
  • Count: a count of events or occurrences. Uses PQL with a Poisson log-link and reports incidence rate ratios (IRR).
What is the ICC and why does it matter?
The Intraclass Correlation (ICC) is the proportion of the total outcome variance that is explained by group membership. An ICC of 0.15 means 15% of the variance is between groups, the rest is within-group individual variation.

As a rule of thumb: ICC > 0.05 justifies a multilevel model. ICC < 0.05 suggests the grouping is not meaningful and a standard regression would give equivalent results. AnalyZ reports the ICC alongside a plain-language interpretation and includes an independent verification panel showing a naive ICC computed directly from group means, so you can confirm the model-based estimate is reliable.
What is a Random Intercept versus a Random Slope model?
Random Intercept: Each group has its own baseline level (intercept), but the effect of every predictor is assumed to be the same across all groups. This is right for most nested datasets, for example, clinics differ in their baseline vaccination rates but the effect of a predictor like distance-to-clinic is the same everywhere.

Random Slope: One predictor is allowed to have a different effect in different groups. For example, the effect of an intervention may be stronger in urban clinics than rural ones. This adds a variance component for that slope. Use it when you have a specific theoretical reason to expect effect heterogeneity across groups.
Does AnalyZ support more than two levels of nesting?
Yes. The Multilevel module supports 2-level, 3-level, and 4-level models: for example, patients nested within clinics nested within districts nested within regions. Select the number of levels and assign a grouping variable for each level. A hierarchy check runs automatically: it tells you whether your data is truly nested (each lower-level unit belongs to only one upper-level unit) or crossed/partially crossed, and flags groups with very few observations.
What are non-parametric tests and when should I use them?
Non-parametric tests are distribution-free alternatives to t-tests and ANOVA. Use them when your outcome variable is ordinal (e.g. a Likert scale), when the normality assumption is clearly violated, or when your sample is small. AnalyZ offers four non-parametric tests:
  • Mann-Whitney U: alternative to the independent t-test for two groups. Reports medians, mean ranks, the U statistic, and rank-biserial r effect size.
  • Wilcoxon Signed-Rank: alternative to the paired t-test for two matched measurements. Reports the W statistic and r effect size.
  • Kruskal-Wallis: alternative to one-way ANOVA for three or more groups. Includes Dunn post-hoc comparisons with Bonferroni correction and eta-squared effect size.
  • Friedman Test: alternative to repeated-measures ANOVA for three or more matched conditions. Includes Wilcoxon pairwise post-hoc tests with Bonferroni correction and Kendall's W.
If your data is roughly normally distributed, the parametric tests (t-test, ANOVA) are more powerful. When in doubt, run both and compare.
What is Repeated-Measures ANOVA and when should I use it?
Repeated-Measures ANOVA tests whether a continuous outcome differs significantly across three or more time points or conditions measured on the same subjects, for example, a knowledge score at baseline, midline, and endline. It is the parametric counterpart to the Friedman test.

The output includes condition means, an ANOVA table with F-statistic and partial eta-squared effect size, a Mauchly's sphericity check, and Bonferroni-corrected pairwise post-hoc comparisons when the overall test is significant. If Mauchly's W is below 0.75, the sphericity assumption may be violated, AnalyZ flags this and recommends interpreting results with caution. If normality cannot be assumed, use the Friedman test instead.
When should I use Poisson versus Negative Binomial Regression?
Both models are designed for count outcomes, non-negative integer values such as the number of services used, incidents reported, or visits made. Results are reported as Incidence Rate Ratios (IRR): an IRR above 1 means the predictor is associated with a higher count rate; below 1, a lower rate.

Poisson Regression assumes the mean and variance of the count are equal. AnalyZ runs an automatic overdispersion check (Pearson chi²/N) and displays a warning if the assumption is violated.

Negative Binomial Regression relaxes this assumption and allows the variance to exceed the mean, common in humanitarian and health data where a small number of cases have disproportionately high counts. The output includes an Alpha parameter measuring the degree of overdispersion; if Alpha is near zero, Poisson and Negative Binomial give similar results.
Effect Analysis
What is Effect Analysis and how is it different from Inferential Analysis?
Inferential Analysis tests whether two variables are related: a correlation, a group difference, a predictor in a regression. Effect Analysis asks a different question: whether a treatment or programme caused a change, using quasi-experimental designs that account for the fact that participation usually wasn't randomly assigned. It offers two methods: Propensity Score Matching (for a single cross-sectional dataset) and Difference-in-Differences (for treatment and comparison groups observed at two time points).
What is Propensity Score Matching and when should I use it?
Propensity Score Matching (PSM) constructs a comparable comparison group from your existing data by matching or reweighting units on their observed characteristics, so treated and untreated units look statistically alike on everything except treatment. Use it for a single cross-sectional dataset with no repeated rounds, where treatment wasn't randomly assigned and you need a credible comparison group built from observed characteristics.
What estimation techniques does PSM support?
Four, selectable per run:
  • Nearest neighbor matching: pairs each treated unit with the most similar control (1:1, within a caliper). Intuitive and transparent, but drops unmatched units.
  • Kernel matching: builds a weighted counterfactual from all controls, weighted by similarity. Keeps every unit.
  • Inverse probability weighting (IPW): reweights the full sample rather than forming explicit pairs.
  • Doubly robust (AIPW): combines a propensity-score model with an outcome regression model; stays valid even if only one is correctly specified.
What's the difference between ATT, ATC, and ATE?
These are the three estimands PSM can report, independent of which technique you choose. ATT (Average Treatment effect on the Treated) is the effect on those who actually received treatment. ATC (Average Treatment effect on the Controls) is what the effect would have been on those who didn't receive treatment. ATE (Average Treatment Effect) is the effect across the full population, combining both.
How do I know if my matching actually worked?
Check the diagnostics AnalyZ reports alongside every PSM result before trusting the treatment effect:
  • Covariate balance table: the standardized mean difference (SMD) for each covariate, before and after matching. |SMD| below 0.1 is the conventional threshold for good balance.
  • Matching quality summary: how many covariates reached good balance and the overall percent bias reduction.
  • Rubin's B and R: standardized bias (target under 25) and variance ratio (target 0.5–2) of the propensity-score distributions themselves.
A statistically significant effect estimated from a poorly-balanced comparison group is not a reliable finding.
What is the sensitivity analysis (E-value) for?
Matching and weighting can only adjust for confounders you actually measured, they can't rule out unmeasured ones. The E-value is the minimum strength of association an unmeasured confounder would need with both treatment and outcome to fully explain away the estimate; larger values mean the finding is more robust. The accompanying bias grid shows how the estimate would shift under a range of hypothetical confounding strengths, so you can see the trajectory rather than just one threshold.
What is Difference-in-Differences and when should I use it?
Difference-in-Differences (DiD) compares how an outcome changed over time between a treatment/programme group and a comparison group. The estimate is the group × round interaction, the change for the treatment group beyond whatever the comparison group experienced over the same period. Use it when you have both a treatment and a comparison group AND outcome data from two time points for both, it rules out secular trends that a simple before-after comparison alone can't distinguish from a real programme effect.
Do I need two separate files for Difference-in-Differences?
Not necessarily. If your uploaded dataset already has a round/period column with both time points in one long-format table, select it directly, no second file needed. If your baseline and endline are separate files, upload the second round from inside the DiD panel and match it to round 1 by an ID column; group and any missing covariates are carried over from round 1 automatically.
What file types can I upload for the second DiD round?
CSV, TSV, Excel (.xlsx/.xls), JSON, Stata (.dta), or SPSS (.sav), the same formats the main sidebar upload supports. The file needs the same column names as round 1 for the ID, outcome, and any covariates.
What is the "parallel trends" assumption in DiD?
Difference-in-Differences assumes the treatment and comparison groups would have followed parallel trends over time if the intervention hadn't happened. This can't be tested with only two time points, AnalyZ flags it as an assumption you're making, not something verified by the analysis. Note also that the subsample filter does not apply when you're using an uploaded second-round file rather than a single long-format dataset.
What is Regression Discontinuity Design and when should I use it?
Regression Discontinuity Design (RDD) compares units just above and just below an eligibility cutoff on a continuous running variable to estimate a treatment effect. Use it whenever programme eligibility is determined by a threshold, a vulnerability score, an income level, an age limit, since units on either side of the cutoff tend to be very similar except for treatment status. It comes with a full diagnostic suite so you can check whether the design is credible before trusting the estimate.
What's the difference between sharp and fuzzy RDD?
Sharp RDD assumes the cutoff perfectly determines treatment, everyone above (or below) the threshold was treated, no exceptions. Fuzzy RDD handles imperfect compliance, where some units above the cutoff weren't actually treated or vice versa. Fuzzy RDD needs an extra variable (the actual treatment received) and is estimated via 2SLS, reporting a Local Average Treatment Effect (LATE) rather than a simple jump in the outcome.
How does bandwidth selection work in RDD?
Only observations within a window (the bandwidth) of the cutoff are used, since the comparison is only credible close to the threshold. You can set the bandwidth manually, or choose automatic, which uses leave-one-out cross-validation (a Ludwig-Miller style method) to pick a data-driven bandwidth balancing bias against variance.
What diagnostics does RDD report, and why do they matter?
A jump at the cutoff is only credible if the design itself checks out. AnalyZ reports four diagnostics with every RDD result:
  • Manipulation / sorting test: checks whether the running variable's density changes discontinuously at the cutoff, which would suggest units sorted themselves around it (e.g. if people learned the threshold and adjusted their reported score).
  • Covariate balance: any covariates you add should show no significant jump at the cutoff. If one does, something besides treatment is changing there too.
  • Bandwidth sensitivity: the estimate re-run at several bandwidths (0.5× to 2× the chosen one). Large swings mean the result is fragile.
  • Placebo cutoffs: the same test re-run at fake cutoffs away from the real one, which should show no significant effect.
Treat the headline treatment effect as provisional until these all come back clean.
What does the "weak first stage" warning mean in fuzzy RDD?
It means the cutoff barely predicts whether a unit was actually treated, the jump in the probability of treatment at the cutoff is small. When that's the case, the fuzzy RD (LATE) estimate, which divides the jump in outcome by the jump in treatment probability, becomes unstable, since dividing by a number close to zero amplifies noise. AnalyZ flags this automatically; check the first-stage table before trusting the estimate.
What is Instrumental Variables (2SLS) and when should I use it?
Instrumental Variables (IV) estimates the effect of a treatment that isn't randomly assigned and may be correlated with unobserved factors, reverse causality, selection, or omitted variables that also affect the outcome, which would bias a plain regression. It uses an instrument: a variable that affects the treatment but has no direct effect on the outcome except through it. AnalyZ estimates this via two-stage least squares (2SLS). Use it when you suspect endogeneity AND you have a credible instrument in mind, the method can't manufacture a good instrument for you.
How do I know if my instrument is any good?
Check the first-stage F-statistic before reading anything else, it measures how strongly the instrument predicts the treatment. F above 10 is the conventional rule of thumb for "not weak." A weak instrument (F below 10) doesn't just widen the confidence interval, it can produce a badly biased point estimate too, however clean the number looks. Beyond strength, the instrument also needs a credible exclusion restriction (no direct effect on the outcome except through treatment), this is a substantive judgment about your programme that no statistic can fully verify.
What's the difference between the OLS, first-stage, and 2SLS results shown for IV?
OLS is the naive estimate that ignores endogeneity, shown for comparison, not as the answer. The first stage is a regression of the treatment on the instrument(s), used to check instrument strength (the F-statistic) and to construct the 2SLS estimate. 2SLS is the actual instrumented estimate, the one that corrects for endogeneity, assuming the instrument is valid. A large gap between OLS and 2SLS, combined with a significant endogeneity test, is evidence the treatment really was correlated with something unobserved.
What are the endogeneity and overidentification tests for?
The endogeneity test (Durbin-Wu-Hausman) checks whether the treatment is actually endogenous, a significant result supports using 2SLS over plain OLS; a non-significant one just means the test couldn't detect a difference in this sample, not that treatment is proven exogenous. The overidentification test (Sargan) only runs when you have more than one instrument, and checks whether the instruments agree with each other. A significant result suggests at least one instrument may not be valid, but passing this test never proves an instrument is good, since a single-instrument (just-identified) model has no equivalent check at all.
Can I use more than one ID column to match rounds in DiD?
Yes. When uploading a second round, select one ID column if it's already unique (a household ID), or select several together if only the combination is unique, for example county + sub-county + serial number, where none of the three alone identifies a unit but the combination does. AnalyZ builds a composite key from whichever columns you select and matches units across rounds on that.
Do the causal methods show standard model-fit statistics like R²?
Yes, tailored to what's actually meaningful for each method rather than the same three numbers everywhere:
  • Difference-in-Differences and Regression Discontinuity Design (sharp) report R², adjusted R², and an F-statistic, standard, since both are ordinary regressions.
  • Instrumental Variables (and fuzzy RDD, which also uses 2SLS) reports first-stage R², a genuine fit measure, since the first stage is plain OLS, plus a 2SLS R² shown for reference only, with an explicit warning that it is not a reliable model-quality indicator.
  • Propensity Score Matching reports McFadden's pseudo-R² for the propensity-score model, how well your covariates separate treated from control units, not a fit statistic for the treatment effect itself.
Can my outcome variable be categorical with more than two groups?
No, across all four Effect Analysis methods (PSM, DiD, RDD, IV), the outcome and any treatment variable must be numeric or binary. A multi-category (3+ level) nominal outcome isn't supported and won't appear as a selectable option in those fields. For a binary outcome, AnalyZ automatically handles the 0/1 encoding, including for string-coded values like "yes"/"no".
Structural Equation Modelling
What can I do in Structural Equation Modelling, and which analysis should I choose?
The module has six analyses. Confirmatory Factor Analysis checks that your questionnaire items measure the factors you expect. Structural Equation Model adds paths between those factors. Compare Groups tests whether a scale or a model works the same way in every group. Path Analysis estimates direct and indirect effects among observed variables. Mediation Analysis tests whether the effect of X on Y runs through one mediator. Counterfactual Mediation does the same when variables are binary or ordered, or when you want covariates, several mediators, a categorical exposure or effects by group.

If you use a questionnaire, start with Confirmatory Factor Analysis: effects between constructs are only trustworthy when the constructs are measured well.
Why is Structural Equation Modelling marked Beta?
The module is new. The methods are established, but it has had less real-world use than the older modules. For findings you plan to publish, check them against another package, and tell us at support@analyz.solutions if something looks wrong.
Which variables can I use, and how are categories handled?
All variables must be numeric. Likert items stored as numbers work well. Where an analysis offers a choice of variable types, All continuous treats items as numeric scores, and Mix of binary, ordinal and continuous treats each variable by its number of different values: 2 is binary, 3 to 10 is ordinal, and more than 10 is continuous.

A categorical predictor or covariate can be set to Categorical, which codes it into one 0/1 variable per category against a reference category you choose, with value labels shown when they exist, or to Continuous, which uses its values as numbers.
Does Structural Equation Modelling use survey weights?
Yes. If you have set a survey weight or survey design in Data View, it is used and the weighted N is reported. Bootstrap intervals resample whole clusters when a design is set. The subsample filter also applies, and only complete cases on the variables in the model are used.
How do I know whether my model fits?
A good fit means CFI and TLI of 0.95 or higher, RMSEA of 0.06 or lower, and SRMR of 0.08 or lower. The chi-square test almost always rejects a model in large samples, so judge fit by the other measures. A suggested change should only be added when it makes sense for the wording of your questions and your theory, and every change you add should be reported.
How many items does a factor need, and how large a sample?
Each factor needs at least two items, and three or more is much safer (a single-factor model needs three). Aim for at least 10 cases per estimated parameter, never fewer than 100 cases in total, and at least 200 for structural equation models, more for complex ones.
My path model fits perfectly. Is that good?
Not necessarily. If every possible path is included, the model fits perfectly by design, so the fit tells you nothing. Leave out paths you expect to be zero, because that is what lets the fit be tested.
What is measurement invariance, and which level do I need?
Measurement invariance checks that a scale measures the same thing in every group. Configural invariance means the same structure in each group. Metric adds equal loadings, which lets you compare correlations and paths across groups. Scalar adds equal item intercepts, which lets you also compare group means of a factor. A level is supported when the fit barely changes from the one before. If a level is not supported, you can allow a few loadings or intercepts to differ (partial invariance), but if many items need to differ, the comparison is not safe. The grouping variable needs 2 to 8 groups with at least 30 cases each.
What is the difference between Mediation Analysis and Counterfactual Mediation?
Mediation Analysis uses regression on numeric variables with a bootstrap interval for the indirect effect and the Sobel test. Counterfactual Mediation simulates what the outcome would be under different exposure and mediator values, so it works with continuous, binary and ordered mediators and outcomes, with covariates, and with an optional exposure-by-mediator interaction. Use Counterfactual Mediation when the mediator or the outcome is binary or ordered, or when you want covariates.
How do I read the direct, indirect and total effects?
The total effect is the full effect of the exposure on the outcome. The indirect effect is the part that works through the mediator, and the direct effect is the rest. An effect is significant when its 95% bootstrap interval does not include zero. Full mediation means a significant indirect effect and no significant direct effect. Partial mediation means both are significant. Suppression means they have opposite signs, so interpret it with care. For a binary outcome, effects are in percentage points. The proportion mediated is unstable when the total effect is close to zero, so report it only when the total effect is clearly different from zero.
Can I have more than one mediator?
Yes, up to four in Counterfactual Mediation. Parallel mediators do not affect each other, and the indirect effect is split by route. A serial model has exactly two mediators, the first affecting the second, and separates the route through the first, through the second and through both. With a binary outcome or interactions, the split between routes can depend on the order of the mediators.
Can I use an exposure with several categories, or compare groups?
Yes, under Effects for in Counterfactual Mediation. A categorical exposure with 3 to 10 categories is compared against a reference category you choose, and each other category gets its own total, direct and indirect effect. The comparisons share one reference and are not adjusted for the number of comparisons. Each group separately estimates the whole model within each group of a variable with 2 to 6 groups, so every path can differ. Each group is compared with a reference group, and each difference has an interval from resamples paired between the two groups.
Does mediation analysis prove that the mediator causes the change?
No. The effects describe causal pathways only if no unmeasured factor affects the exposure, the mediator and the outcome together. Include the covariates that could do this, and use the sensitivity panel to see how much hidden confounding your result could withstand.
What is the sensitivity analysis in Counterfactual Mediation?
Open Sensitivity to unmeasured confounding above the interpretation. It shows E-values for the total, direct and indirect effects: the smallest strength of link that a hidden factor would need with both sides of the relationship, beyond your covariates, to explain the effect away. Larger values mean the result is harder to explain away. With one continuous mediator and a continuous outcome, it also plots the indirect effect against an assumed correlation between the unmeasured factors behind the mediator and the outcome. The panel shows how robust a result is. It does not remove confounding.
What can I export from Structural Equation Modelling?
Results export to Word and Excel with all tables and charts, and to CSV. Add to Canvas lets you choose which chart to send to your Data Canvas, and Add to Sense Making sends the tables and interpretation. Diagrams can be downloaded as PNG or SVG.
Data Visualisation
What chart types are available?
28 chart types including Bar, Stacked Bar, Histogram, Scatter, Line, Sparkline, Boxplot, Violin, Heatmap, Bubble, Treemap, Pie, Donut, Sunburst, Sankey, Forest plot, Table and Multiple Response (for select-all-that-apply questions). All charts are interactive (zoom, pan, hover for values) and can be exported as PNG, JPG, or interactive HTML.
Can I customise the charts?
Yes. Each chart has a 🎨 Edit chart panel where you can change the title, axis labels, font size, text colour, bar/marker colour, background colour, and axis ranges. Click Apply changes to update the chart. Changes persist for the session.
Can I build a dashboard with multiple charts?
Yes. From Data Visualization, select Multi-chart Dashboard from the opening card screen. Build your charts and they assemble into a responsive grid with cross-filtering: clicking any bar, slice, or data point filters all other charts simultaneously. Building and saving a dashboard privately works on every plan, including Free. Publishing it as a live public link is a Starter/Pro feature, see the next question. The dashboard can also be downloaded as an interactive HTML file on any plan.
Can I share my dashboard with others?
Yes, on Starter and Pro plans. After building a multi-chart dashboard, click Share dashboard. This publishes it as a live link, anyone can open and interact with it in their browser with no AnalyZ account required. You can revoke or update the link at any time from the dashboard view. For a narrative format, see Data Canvas below, a storytelling tool that also publishes as a public link on Starter/Pro.
Data Canvas
What is Data Canvas?
Data Canvas is a storytelling module that turns your analysis outputs into a shareable, scrollable data story powered by AnalyZense. You collect findings from any analysis module using Add to Canvas, then ask AnalyZense to generate a story, it reads all your findings, decides the best way to present them, writes plain-language headlines and narratives, and assembles them into a structured slide story. You can refine individual slides, adjust heights, and publish as a public link anyone can read in their browser with no account required. Designed for communicating findings to non-technical audiences such as donors, community members, or board members.
How is Data Canvas different from the dashboard?
They serve different purposes. The dashboard is an analytical tool, multiple charts side-by-side with cross-filtering, designed for exploring and interacting with data. Data Canvas is a communication tool, a scrollable narrative with insight headlines and plain-language context, designed for telling the story of what the data means to a non-technical audience. You can use both: explore with the dashboard, then add key charts to a Data Canvas story.
What can I add to a Data Canvas?
Every analysis module has an Add to Canvas button. Supported finding types: Chart (any Plotly chart from any analysis module), Table (frequency tables, crosstabs, test statistics), Infographic (network maps and SVG diagrams from qualitative analysis), and Quote (coded participant excerpts from Thematic Coding). These collected findings form your Findings tab: the evidence repository. AnalyZense then reads them all and generates the story.
What does AnalyZense do in Data Canvas?
AnalyZense does two things in Data Canvas. First, Generate Story: you answer three questions (what the study is about, who the audience is, and any key message you want to convey), choose a presentation duration (5, 15, or 30 minutes), and AnalyZense reads all your collected findings, decides the best grouping and layout for each slide, writes new plain-language headlines and narratives, and assembles the full story. Second, Refine slide: after generation, each slide has a Refine button where you give a plain-language instruction (e.g. "make the headline shorter", "convert the forest plot to a bar chart", "lead with the network diagram"). AnalyZense updates just that slide. Both actions count against your monthly allowance shown in the canvas header.
Can I share a Data Canvas story publicly?
Yes, on Starter and Pro plans. Canvases are private by default. When your story is ready, click Publish in the canvas header, a dialog opens where you can write an optional note for readers, and the story is published as a public link. After publishing, the header shows Copy link to copy the URL. If you regenerate or refine slides after publishing, click Re-publish to update the public version. Anyone with the link can read the story in their browser, no AnalyZ account needed, no software required. The viewer is mobile-friendly.
Can I add qualitative findings, like participant quotes, to a Data Canvas?
Yes. The Thematic Coding module has an Add to Canvas button on every coded segment. Clicking it opens the canvas modal where you choose the destination canvas. The quote is added as a styled slide with the excerpt and transcript source as attribution. Qualitative analysis outputs, co-occurrence tables, comparative charts, saturation curves, can also be added as table or chart slides from their respective modules.
Scale & Reliability
What is Cronbach's Alpha and when should I use it?
Cronbach's Alpha measures the internal consistency of a set of scale items, i.e. how well they all measure the same underlying construct. Use it when you have a multi-item scale (e.g. a 5-item anxiety questionnaire) before computing a composite score. A value of 0.70 or above is generally considered acceptable; 0.80+ is good; 0.90+ is excellent.
How do I handle reverse-coded items in Cronbach's Alpha?
Click "Check items for reverse coding" before running. AnalyZ inspects inter-item correlations and flags any item that correlates negatively with the majority, these are likely reverse-scored items. Pre-ticked items are suggested for reversal; confirm or adjust, then click Run.
What is the difference between PCA and EFA?
PCA (Principal Component Analysis) reduces a large set of variables into fewer uncorrelated components that capture the most variance, useful for data simplification or removing multicollinearity. EFA (Exploratory Factor Analysis) models the latent factors that cause variables to correlate, useful for understanding the construct structure of a scale. PCA is a mathematical transformation; EFA is a theoretical model.
What rotation method should I use in EFA?
Use Varimax if you expect your factors to be independent (orthogonal rotation). Use Oblimin if you expect your factors to correlate, for example, "stress" and "burnout" are theoretically related. If unsure, run Oblimin first and check the factor correlation matrix: if all correlations are below 0.32, Varimax is equally appropriate.
Sampling
What does the Sampling module cover?
Sampling brings together three planning tools in one place: Sample Size Estimation (how many participants you need), Power Analysis (the probability your study will detect a real effect, or the sample size needed for a target power), and Draw a Sample (upload a sampling frame and have AnalyZ actually select your sample using your chosen design). All three run before or independently of any loaded dataset.
What study designs does the sample size calculator support?
Calculators are grouped by study design rather than sampling technique. Cross-sectional is one flexible calculator covering a single round of data collection, choose whether you're estimating a proportion or a mean, and how you'll select participants (simple random, stratified, or cluster/multi-stage, which also covers PPS). Measuring change covers Longitudinal Study, Change in Proportion, and Randomised Controlled Trial (RCT), designs built around detecting a difference. Other study designs covers Case-Control and LQAS.
How do I calculate the sample size I need?
Open Sampling, choose Sample Size Estimation, and pick the study design that matches your plan. For Cross-sectional, also choose what you're measuring (a proportion or a mean) and how you'll select participants. Enter your confidence level (typically 95%), margin of error or desired power (typically 80%), and any design-specific parameters. Click Calculate sample size. The result shows the required number with an interpretation and the formula used.
What if I don't know the expected effect size or proportion?
For proportion-based calculators, use 50%, this gives the most conservative (largest) sample size estimate. For power-based calculators, use a medium effect size (Cohen's d = 0.50 for t-tests). You can also base your estimate on results from similar published studies.
Does AnalyZ support estimating a mean, not just a proportion?
Yes. In the Cross-sectional calculator, switch to "Estimating a mean (numeric)" for outcomes like income, age, test scores, or expenditure. You'll need an estimate of the standard deviation, ideally from a pilot study, since the result is sensitive to this assumption.
What is a design effect (DEFF) and how do I set it?
The design effect (DEFF) is a multiplier that accounts for the fact that people within the same cluster (e.g. the same village or school) tend to be more similar to each other than to people in other clusters. A DEFF of 1.5 means you need 50% more participants than a simple random sample would require. It defaults to 1 (no clustering effect) in every calculator and is always editable. When you choose "Cluster / multi-stage" sampling in the Cross-sectional calculator, you can instead calculate DEFF from the intraclass correlation (ICC): DEFF = 1 + (m−1)ρ, where m is your average cluster size. In field surveys, DEFF typically ranges from 1.5 to 3.0.
How does stratified sampling work, and what if I need precision for each subgroup?
Choose "Stratified" as your sampling approach in the Cross-sectional calculator, then add each stratum with a name and population size. By default, the total sample is split across strata proportionate to population size. If you need to report or compare results by subgroup, check "I need this precision within each stratum", every stratum then gets the full sample size independently, since a large national total can still leave individual strata underpowered.
How does the longitudinal calculator handle attrition?
The longitudinal calculator asks for the number of waves and expected attrition per wave (% of participants expected to drop out between each round). It calculates the number of participants you need to enrol at baseline so that enough remain by the final wave for valid analysis. The formula is: enrol = n_required / (1 − attrition)^(waves − 1).
Should I use the independent-groups or paired design in Change in Proportion?
Use two independent groups when comparing two separate samples, for example, a treatment group vs. a control group, or two different groups surveyed at each time point. Use same group, before/after when the same individuals are measured at both time points. The paired design typically needs a smaller sample for the same effect, since each participant acts as their own comparison, you'll also enter an expected correlation between the two measurements (0.5 is a reasonable default), and a higher correlation further reduces the required sample.
What is statistical power and why does it matter?
Power (1−β) is the probability that your study will detect a true effect if one exists. A power of 0.80 means an 80% chance of getting a significant result when the effect is real. Studies with low power miss real effects and produce unreliable results. Most journals and ethics boards require power justification, typically 80% or 90%.
What tests are available in Power Analysis?
Two-sample t-test, one-sample t-test, paired t-test, chi-square test, McNemar's test, one-way ANOVA, linear regression, and logistic regression. Each supports two modes: enter a sample size to see the achieved power, or enter a target power to find the required sample size. A power curve chart shows how power changes across a range of sample sizes for small, medium, and large effects.
Should I use a paired test or a two-sample test?
Use a paired t-test or McNemar's test when the same individuals are measured twice, for example, a knowledge score before and after training. Use a two-sample test or chi-square test when you are comparing two independent groups, such as a treatment group versus a separate control group. Paired designs generally require fewer participants for the same effect size.
What is the Events Per Variable (EPV) rule in logistic regression?
EPV is the number of outcome events divided by the number of predictors in your model. A conventional minimum of 10 events per predictor helps avoid an unstable or overfit model. The Power Analysis logistic regression calculator reports the required sample size from both the power calculation and the EPV rule, and uses whichever is larger.
What is Draw a Sample and how does it work?
Draw a Sample lets you upload your sampling frame (a CSV or Excel file listing every unit in your population) and have AnalyZ select the sample for you. Choose a design, Simple Random, Systematic, Stratified, Cluster, or Probability Proportional to Size (PPS), set your parameters, and click Draw sample. The selected units are shown in a preview table and can be downloaded as CSV or Excel, either as the selected rows alone or the full frame with a "selected" flag column.
Is my sampling frame uploaded to AnalyZ's servers?
No. Like all data in AnalyZ, your sampling frame is processed entirely in your browser. It is never uploaded to or stored on our servers.
Can I reproduce the same sample later?
Yes. Draw a Sample includes an optional random seed. Setting a seed before drawing means the same seed, frame, and design will always reproduce the exact same sample, useful for audit trails or re-generating a lost sample list. Note that a seed is specific to AnalyZ's random number generator and will not reproduce the same draw in other software such as R or Stata.
Qualitative Analysis
What qualitative file formats are supported?
Plain text (.txt), Word documents (.docx), PDF (.pdf, text is extracted automatically) and CSV (.csv). A CSV that looks like a table of numbers is flagged as a dataset instead. You can upload up to 10 files at once (up to 250 MB total). Registered users only, Guest mode is limited to 1 file / 1 MB.
How does thematic coding work in AnalyZ?
Open Qualitative Analysis → Thematic Coding → Assign Codes. The transcript appears on the left. Click and drag to highlight the passage you want to code. With text selected, click any code button in the Coding Panel on the right, the segment is saved immediately with that code and highlighted with a coloured underline. You can assign multiple codes to the same passage and add an optional analyst note before clicking a code.
How do I build a codebook?
Go to Thematic Coding → Manage Codes. Click ➕ Add new code, enter a name, choose a colour, and optionally add a description. Codes can be edited or deleted at any time, deleting a code removes it from all coded segments automatically. You can also add new codes on-the-fly from within the Assign Codes workspace.
What visualisations are available for qualitative data?
Explore Themes (4 cards): Saturation (new codes per transcript), Theme Comparison (per transcript or by participant group), Theme Connectedness (co-occurrence and distinctiveness) and Segment Length. Analyze Relationships (6 cards): Network Map, Thematic Clusters, Code Sequence (Sankey), Configurational Analysis, Qualitative Comparative Analysis (QCA) and Persona Builder. Word cloud and word search are in View Transcripts. Every view includes an automated plain-language interpretation.
What is the Persona Builder?
Persona Builder (Analyze Relationships) groups participants into thematic personas using the co-occurrence patterns in your coded segments. You select the codes that define the thematic space, review a suggested number of personas (solutions from 2 to 8 are tested and the most distinct is recommended), then name and review each one. Personas are data-driven and described by their dominant themes, not generated by AI.
What is Qualitative Comparative Analysis (QCA)?
QCA (Analyze Relationships) is a crisp-set method that finds which combinations of codes (conditions) are linked to an outcome. You select the conditions, assign a binary outcome to each transcript, review the truth table with consistency scores, and read the minimised Boolean solution. The consistency threshold is adjustable (default 0.75).
What does Configurational Analysis show?
It shows a presence/absence matrix of codes per transcript, sorted so transcripts with identical thematic profiles sit together. Each distinct profile is a participant type, and codes are classed as Universal, Differentiating or Rare. Use the Min codes per transcript slider to leave out lightly coded transcripts.
What do Theme Connectedness and Segment Length show?
Theme Connectedness (Explore Themes) has two views: Co-occurrence, how often code pairs appear together, and Distinctiveness, which code pairs rarely or never appear together. Segment Length shows the average word count of coded segments per code, a rough proxy for how much participants elaborate on a theme.
How do I compare themes across participant groups?
Open Explore Themes → Comparative. The tab has two modes:
  • Per transcript: always available, no setup needed. Shows each code's share of total code assignments per transcript as a heatmap and raw count table. Useful for seeing which themes dominated each individual interview.
  • By participant group: requires tag categories. Use Transcript Settings (under View Transcripts) to define categories (e.g. Gender: Male/Female, Location: Rural/Urban) and assign each transcript a value. This mode then shows normalised code frequencies broken down by each group, enabling fair comparison regardless of group size.
Can I save my qualitative coding work?
Yes. Click Save Session in the sidebar to download a .analyz.json file. This saves all transcripts, your full codebook (codes, colours, memos, and keywords), and every coded segment with its assigned codes and analyst notes. Use Restore Session anytime to resume, all coding work is restored automatically.

Save on this device (the toggle in the sidebar) keeps a working copy in your browser's local storage that survives page refreshes within the same session. It expires after 24 hours and is not restored after fully closing and reopening the browser. It is not a substitute for a downloaded session file.
How does Compare Coding work?
Compare Coding is a standalone module for inter-rater reliability. Each coder exports their coding using the JSON (IRR) button in Review Codes (inside Thematic Coding). Load both exports in Compare Coding, give each a label, and click Compute agreement score. The module computes Cohen's Kappa overall and per code, categorising each as strong (κ ≥ 0.8), acceptable (0.6–0.8), or below threshold (<0.6). The interpretation panel gives specific next steps for low-agreement codes.
Sense Making
What is Sense Making?
Sense Making is an AI-assisted feature that synthesises the analyses you have run into a single coherent interpretation. Instead of reading each result separately, the AI connects your findings, identifying patterns, contradictions, and key takeaways across quantitative and qualitative analyses together.
How does Sense Making work?
  1. Run any analyses across any modules.
  2. On each result, click ➕ Add to Sense Making to promote it to a project.
  3. Go to Sense Making in the sidebar and open your project.
  4. Optionally add context about your research question and audience.
  5. Choose an interpretation mode and click ✨ Interpret project.
  6. The AI streams its interpretation. You can then refine it with a plain-language instruction, the interpretation is rewritten and replaces the current version.
What is an AnalyZense project?
A project is a named collection of analyses you have chosen to include in a synthesis. You can have multiple projects, for example, one per research study. Each project stores up to 20 analyses, an optional context description, and your saved interpretations. Projects are private to your account.
What are the interpretation modes?
There are four modes, selectable per project:
  • 🗣️ Plain language: simple words, no jargon, suitable for any audience
  • 🎓 Academic: formal tone, precise statistical language, suitable for research reports and papers
  • 💼 Executive summary: concise, action-oriented, focused on implications for decision-makers
  • 📖 Narrative: flowing prose, suitable for donor reports or public-facing outputs
You can switch modes and re-interpret at any time.
Does the AI see my raw data?
No. The AI only ever sees what you have explicitly added to your project, aggregated result tables, key statistics, auto-generated interpretation text, and any notes you wrote. Your raw dataset is never sent to the AI or stored anywhere.
How many analyses can I add to a project?
Up to 20 analyses per project. You will see a warning at 15 and a hard block at 20. If you need more, remove less relevant analyses or create a new project. You can also copy an analysis from one project to another.
Can I refine the interpretation after it is generated?
Yes. After the initial synthesis, a Refine interpretation panel appears. Type a plain-language instruction, for example, "make this more concise", "focus more on the qualitative findings", "include more participant quotes", or "sharpen the conclusion", and the interpretation is rewritten to incorporate it. The refined version replaces the original. Refining counts as one interpretation from your monthly allowance. You can refine as many times as you like within your limit.
Can I export the interpretation?
Yes. Click ⬇ Export to Word in the synthesis view. The exported document includes the project name, interpretation mode, context, and the full AI interpretation.
Can I control how long the Sense Making interpretation is?
Yes. Each mode has a length setting (Short, Standard or Detailed, except Executive summary which has Short and Standard). Targets range from about 200 words for a short executive summary to about 2,000 for a detailed academic one.
Can I get an earlier version of my interpretation back?
Yes. Open Version history in the project workspace, find the version you want and click Restore. Re-interpreting keeps the previous version in the history, and a cut-off response is never saved or counted.
Can I read a Data Canvas story slide by slide?
Yes. Use the toggle in the canvas header to switch between Vertical scroll and Slide-by-slide. The setting is saved with the canvas. You can also click Edit text on any slide to change the headline and narrative directly.
How many Sense Making interpretations can I run per month?
On the free plan: 1 interpretation per calendar month (each refinement also counts as one interpretation). These reset on the 1st of each month. Your remaining allowance is shown at the top of each project workspace. Paid plans (Starter at ₹200/month and Pro at ₹400/month) offer 10 and 50 interpretations per month respectively.
Pricing & Plans
What is included in the free plan?
Full quantitative analysis, full qualitative analysis, plain-English interpretations at every step, all export formats, session saving and restoring, all free. Monthly AI limits apply: 1 Sense Making interpretation, 1 dashboard interpretation, 2 Canvas story generations, and 5 Canvas slide refinements per month. Sharing (public links for dashboards and Data Canvas stories) requires a paid plan. No credit card required, no time limit.
What paid plans are available?
Feature Free Starter ₹200/mo Pro ₹400/mo
Sense Making 1/month 10/month 50/month
Dashboard interpretations 1/month 10/month 50/month
Canvas: Generate Story 2/month 5/month 15/month
Canvas: Refine Slide 5/month 15/month 50/month
Sharing (Dashboard, Canvas public links) No ✓ ✓
Priority support No ✓ ✓
Early access to new features No No ✓

Enterprise plans with team access and custom limits are available, email support@analyz.solutions.
How do I upgrade my plan?
Self-service upgrades are available. From the home screen open Account Settings, go to the Subscription tab, and click Subscribe on Starter or Pro (or Upgrade to move from Starter to Pro). Payment opens in a secure Razorpay checkout, billed monthly in Indian rupees (the website may show an approximate price in your local currency, but you are billed in INR). You can cancel anytime. Your plan limits update once the payment is confirmed, and the tab shows your renewal date. Plans renew automatically every month. Each subscription covers 12 monthly payments, after which it ends and your account returns to the Free plan unless you subscribe again. If you upgrade from Starter to Pro, the Starter subscription is cancelled automatically so you are not billed twice. If you need an invoice, a different currency or team access, email support@analyz.solutions.
How do I cancel or manage my subscription?
You can cancel anytime. Open Account Settings from the home screen, go to the Subscription tab and click Cancel plan, then confirm. For any other billing change, email support@analyz.solutions. After cancellation your plan stays active until the end of the period you have already paid for (shown as "Access until" in Account Settings), then returns to Free. No partial refunds are issued. If a payment fails and the subscription is paused, your account returns to the Free plan limits straight away. Your data, sessions and saved work are not deleted.
When does my monthly allowance reset?
On the 1st of each calendar month (UTC). Your remaining allowance is always shown in the Sense Making workspace so you know exactly where you stand.
Do unused interpretations roll over to the next month?
No, unused allowance does not roll over. The counter resets to the full monthly limit on the 1st of each month regardless of how many you used the previous month.
Export & Support
How do I export my results?
Every analysis module has export buttons below the results. Tables download as Excel (.xlsx) or Word (.docx). Charts export as PNG, JPG, or interactive HTML. Multi-chart dashboards can be published as a shareable live link (viewable by anyone, no login needed) or downloaded as interactive HTML. Data Canvas stories can also be published as a public link, a scrollable narrative with charts, tables, and quotes. The sidebar offers dataset download in CSV, Excel, Stata, or SPSS format. In Sense Making, click ⬇ Export to Word to download the full AI interpretation and context as a formatted Word document.
How do I get support?
Email us at support@analyz.solutions. We aim to respond within 48 hours on business days.
Where can I find detailed instructions for each module?
The full User Guide covers every module with step-by-step instructions, reference tables, and tips. It is accessible any time from the Help & FAQ section inside the app or from the Support section of the landing page.