Everything ZStat can do

Arranged in the order the work happens, not alphabetically. You are on the Free plan.

Statistics The analysis itself Machine learning Prediction, not explanation Data collection From questionnaire to responses Writing up The chapter, not the output Industry and quality Manufacturing, process, operations Planning and the built environment Urban planning, estate management, architecture Ethics and privacy NDPA 2023, enforced What do I use? By field, in your words Questions Answered plainly

Every procedure, searchable

Type what you are trying to do. The catalogue is written in those words rather than in procedure names.

What a plan never limits. Every assumption check, every refusal, every leakage guard and every compliance blocker is on all plans including the free one. A plan limits how much you can do, never how correctly.

Three of these decisions cannot be revisited once collection starts. Most of what goes wrong in a thesis goes wrong here, months before anyone looks at a number.

design.advise advisor

Which research design can answer your question — and, more usefully, what each one cannot support

design.sampling advice advisor

Whether your sampling method supports the claim you want to make

design.what comes next advisor

Where you are in the process and which stages are already locked

power analysis procedure

How many participants you need to detect an effect worth detecting

instrument health check

Whether your questionnaire is fit to lock: double-barrelled items, leading wording, reading level

demo.build reference

A complete synthetic study to look at before you have data of your own

examples/ddos ensemble.py reference

A worked intrusion-detection study at 70/15/15, grouped by capture session

demo.walkthrough reference

Six things to try in five minutes, each with why it matters

collection readiness check

Whether you may lawfully collect yet - ethics approval, consent notice, NDPA 2023 compliance

Cleaning and feature engineering are different jobs that go in different sections of the thesis and carry different risk. Keeping them apart is why they are listed separately here.

interop.spss metadata io

What an SPSS file already knows, before importing it

sample.rows io

Fewer rows - random, stratified or systematic - with a sentence for the methods section

sample.columns io

Fewer columns, with what was dropped recorded

io.read io

Open CSV, Excel or SPSS, detecting the encoding, the separator and the missing codes

prepare.apply missing cleaning

Treat a code such as 99 as missing rather than as data

prepare.recode cleaning

Correct or collapse category values

prepare.flag outliers cleaning

Mark extreme values - flagged, never removed

prepare.filter rows cleaning

Remove cases, with the reason recorded

missing pattern procedure

Whether the gaps are random, or related to what people answered

prepare.scale score engineering

Combine items into a scale score, with reverse-scoring

prepare.transform engineering

Log, square root, rank or z-score

prepare.bin values engineering

Cut a continuous variable into groups

The commonest question in a thesis, and the one where the wrong test is most often used.

exact test procedure

Exact binomial, McNemar and runs tests, by enumeration

multidimensional scaling procedure

Placing objects so distance reproduces dissimilarity

one sample t procedure

One group against a known value

independent t procedure

Two separate groups

paired t procedure

The same people twice

one way anova procedure

Three or more groups

factorial anova procedure

Two grouping variables and their interaction

ancova procedure

Groups compared with a covariate held constant

repeated measures anova procedure

The same people three or more times

manova procedure

Several related outcomes at once

mann whitney procedure

Two groups, skewed or ordinal data

kruskal wallis procedure

Three or more groups, skewed or ordinal

wilcoxon signed rank procedure

The same people twice, skewed or ordinal

friedman procedure

The same people three or more times, ordinal

sign test procedure

Paired data using only the direction of change

permutation test procedure

An exact p value with no distributional assumption

bootstrap difference procedure

The difference between two groups, with an interval, on skewed data

mcnemar procedure

A yes/no outcome measured twice on the same people

chi square independence procedure

Two categorical variables

chi square goodness of fit procedure

Whether one categorical variable matches an expected distribution

fisher exact procedure

A small contingency table

survey crosstab procedure

A contingency table under a complex sample design

loglinear procedure

Three or more categorical variables at once

correspondence analysis procedure

A map of which categories go with which

multiple correspondence procedure

The same, for three or more variables

nonlinear canonical procedure

Relating two sets of categorical variables

discriminant procedure

Which variables separate known groups

probit procedure

A binary outcome through the normal curve

curve estimation procedure

Several curve shapes fitted and compared

weighted least squares procedure

Regression where some cases are measured more precisely

two stage least squares procedure

When a predictor is itself caused by the outcome

survey regression procedure

Regression under a complex sample design

cart procedure

A classification tree, cross-validated

quest procedure

A tree that picks the variable before the cut point

neural network procedure

MLP or RBF, compared against logistic regression

rfm procedure

Recency, frequency and monetary scoring

pearson r procedure

Two continuous variables

spearman rho procedure

Two variables, ranked

kendall tau procedure

Two ordinal variables, small sample

partial correlation procedure

Two variables with a third held constant

correlation matrix procedure

Everything against everything, with the multiplicity handled

ols regression procedure

Predicting a number from several predictors

logistic regression procedure

Predicting yes or no

multinomial logistic procedure

Predicting one of several categories

morans i procedure

Whether high values on a map sit next to other high values — testing it by permutation rather than a normal approximation that assumes the independence under test

getis ord procedure

Which specific places are hot spots or cold spots — saying how many would look significant by chance alone, because one test is run per location

nearest neighbour procedure

Whether points are clustered, random or evenly spaced — naming where the study area came from, because the answer depends entirely on it

control chart procedure

Whether a process is stable over time — refusing to let control limits be replaced by specification limits, which hides the drift the chart exists to show

process capability procedure

Whether a process can meet its specification — checking stability first, because a capability index describes a distribution an unstable process does not have

measurement invariance procedure

Whether a questionnaire means the same thing to different groups — refusing to endorse a comparison of mean scores when the intercepts differ, whatever a t test on the totals would say

irt 2pl procedure

How hard each item is and how well it separates people — saying when the sample is too small to support the item parameters it just produced

network analysis procedure

Which items relate to which, holding the rest constant — stating that a single network's centrality ordering is unstable and needs a bootstrap before it is interpreted

roc comparison procedure

How well a test tells cases from non-cases — comparing two tests on the same patients with DeLong's method, which accounts for the correlation that an ordinary comparison ignores

propensity matching procedure

Comparing treated and untreated people fairly — reporting covariate balance before and after, and refusing when the two groups are too dissimilar to match

non inferiority procedure

Showing a treatment is not worse by more than a margin — requiring the margin to be stated, because a non-significant superiority test is not evidence of equivalence

panel regression procedure

Data on the same firms or people over several periods — running the poolability F, Breusch-Pagan LM and Hausman tests so the choice between fixed and random effects is justified rather than asserted

mediation procedure

Does X affect Y through M? Bootstrapped indirect effect, not Baron and Kenny

moderation procedure

Does the effect of X depend on W? With simple slopes

multilevel empty procedure

Whether the grouping matters at all, before modelling it

multilevel mixed procedure

Prediction where cases are grouped in schools or wards

ordinal logistic procedure

Predicting an ordered outcome such as a rating

gee procedure

Repeated measures, population-average effects

sem procedure

Latent variables and paths between them

chaid procedure

Which combination of categories separates the outcome

conjoint procedure

What people are trading off between attributes

The models are the easy part. Everything here is about the ways a result can look excellent and be worthless.

ml.classify ml

Predict a category, with seven models compared against a baseline

ml.estimate ml

Predict a number

ml.text classify ml

Predict a category from written text

ml.group ml

Find clusters

ml.detect anomaly ml

Find unusual cases when you cannot label

ml.multi label ml

Several labels per case

ml.survival ml

Time until an event

dl.images ml

Image classification, distillation, compression robustness

ml.ensembles ml

Voting and stacking, with how often the members disagree

ml.imbalance ml

What to do about a rare class, and what none of it fixes

ml.signals ml

Features from a waveform: RMS, crest factor, kurtosis, spectral bands

ml.conformal ml

Prediction intervals with a guaranteed coverage rate

ml.explain case ml

Why this particular case, not just which variables matter overall

ml.save ml

Save a model with its known faults travelling inside

ml.audit check

Calibration, subgroup performance and cost-sensitive thresholds

ml.guards check

Leakage, imbalance and thin data, refused before training

ml.graph text check

Transductive versus inductive, campaign duplication, homophily

Evaluating ZStat for an institution? The data protection and architecture assurance is generated from this running installation and can be downloaded as a Word document for your ICT directorate or procurement file.

Plans

ZStat comes two ways. They are different products with different obligations, not two prices for the same thing.

Hosted by ZStat

We run it. Individuals start free; departments and institutions get their own space with their own subdomain, or point their own domain at us.

Where the data sits	On ZStat infrastructure. Each institution has its own database; individual accounts share a pooled one.
Who is the controller	The institution, or the individual researcher, is the data controller. ZStat is the processor.
Bought	On this site

Free yours

No charge, no time limit

Designing a study, piloting an instrument, and learning what your data will and will not support. Enough to do the work; not a substitute for buying the tool.

Studies	1
Responses per study	150
Members per study	1
Stored analyses	100
Ml rows	5,000
Image models	no limit

All 86 statistical procedures
Every assumption check and refusal
Questionnaire designer and health check
NDPA-compliant collection, up to 150 responses
Every result and figure shown in full on screen
CSV export of any table
Machine learning on tabular and text data
not included: Word and PDF export without a footer line
not included: More than 150 responses in a study
not included: Supervisor or co-investigator access
not included: Verification certificates
not included: The offline desktop application

Researcher

Per year, per person

Someone running studies continuously - a lecturer, a consultant, a research officer.

Studies	25
Responses per study	20,000
Members per study	10
Stored analyses	10,000
Ml rows	500,000
Image models	50

Everything in Free
Up to 25 studies
ZStat Studio for Windows, offline
Image models and knowledge distillation
Portable packages for offline fieldwork
Email support
not included: Institution-wide administration

If the licence lapses: read-only. You keep signing in, reading every result, exporting your data and reports, honouring withdrawals and running retention. New studies, collection and analyses pause until you renew.

Institution

Per year, per institution

A department or university running many studies, with supervision and record-keeping across all of them.

Studies	no limit
Responses per study	no limit
Members per study	no limit
Stored analyses	no limit
Ml rows	no limit
Image models	no limit

Everything in Researcher
Unlimited studies
Your own subdomain, or your own domain pointed at us
Your institution's data in its own database
Institution-wide audit and oversight
Named data protection officer per study
Training and priority support
not included: Running it on your own hardware - that is ZStat Self-Hosted, sold separately

Department

Per year, per department

One department or faculty, without waiting for a university-wide decision.

Studies	60
Responses per study	20,000
Members per study	15
Stored analyses	20,000
Ml rows	500,000
Image models	100

Everything in Researcher, for everyone in the department
A named unit inside your institution on ZStat
Departmental audit and oversight
Supervisors and co-investigators included
Upgrades to Institution without moving any data
not included: A subdomain of your own - that is the Institution plan
not included: Running it on your own hardware

Installed on your own server

You run it. Respondent data never reaches us, and we cannot see it, produce it or delete it.

Where the data sits	Entirely on your infrastructure.
Who is the controller	You are the controller and the processor. ZStat is a software supplier and nothing more.
Bought	https://ztech.ng

Self-Hosted

Per year, per institution, licensed

A university that keeps respondent data on its own hardware, under its own data protection officer and inside its own network.

Studies	no limit
Responses per study	no limit
Members per study	no limit
Stored analyses	no limit
Ml rows	no limit
Image models	no limit

Everything in Institution
Runs on your server, inside your network
Respondent data never leaves your infrastructure
Your own backup and retention regime
Annual licence, renewed on ztech.ng
Installation support and upgrade releases
not included: Hosting, monitoring and backups - those are yours
not included: We cannot recover data we do not hold

Why a lapsed licence never locks you out. A lapsed licence stops new work. It never withholds data or the ability to export it. Under NDPA 2023 the account holder is the data controller and remains legally obliged to produce, correct and delete that data on request - software that blocks them has manufactured a compliance breach for its own customer.

On every plan, including the free one

Assumption checks	Every test reports what it assumed and whether that held.
Refusals	A test that cannot answer the question is refused, with the reason, on every plan.
Leakage guards	Machine learning refuses data that would produce a meaningless result.
Compliance blockers	Collection is blocked until ethics approval and the consent notice are in place. NDPA 2023 applies to a free user exactly as it does to a paying one.
Instrument health	A questionnaire with a double-barrelled or leading item cannot be locked.
Withdrawal	A respondent can always withdraw, and their data always leaves the analysis.
Audit log	Everything that happens is recorded and cannot be edited.
Provenance	Every result carries a hash of the data it came from.
All procedures	All 79 statistical procedures. Which test is right is a matter of the design, not of what was paid.
Seeing your results	Every result, table and figure is shown in full on screen and can be copied. What a plan buys is the formatted document, not the finding.

Research tips by email

Short, specific methodological notes: each one names a mistake worth avoiding and what to do instead. Weekly or daily, one click to stop, and no tracking of whether you open them.