Analyses

Stationarity tests

Augmented Dickey–Fuller and KPSS tests on one numeric series.

Edit on GitHub

When to use it

Use these tests when you need to know whether a series looks stationary before ARIMA. ADF’s null is a unit root. KPSS’s null is stationarity around a constant. The two can disagree; that disagreement is the result to read. Autocorrelation is the other diagnostic.

Assumptions

The target is numeric, with at least eight non-missing values. ADF uses AIC lag selection. KPSS uses a constant (level stationarity) and automatic lags. There is no trend-stationarity option and no chart.

Running it in Tensr

Time series → Decomposition → Stationarity Tests. In chat: “ADF and KPSS of sales.”
Target is the numeric series. Date is optional.

Options

Prop

Type

Reading the output

The same sales series, which was built with a drift and a seasonal wiggle, not as white noise. ADF’s null is a unit root. KPSS’s null is level stationarity.

Value
Statistic-1.7502
p-valuep = .405
Critical 1%-3.6155
Critical 5%-2.9413
Critical 10%-2.6092
Value
Statistic0.3558
p-valuep = .096
Critical 10%0.3470
Critical 5%0.4630
Critical 2.5%0.5740
Critical 1%0.7390

ADF p = .405 (null: unit root). KPSS p = .096 (null: level stationarity). Neither test rejects its null on this short, mildly drifting series, so the pair is inconclusive.

Reporting (APA 7)

ADF p = .405, KPSS p = .096 on 48 months of sales. Report both tests; they do not share a null.

Coming from SPSS

SPSS does not ship ADF and KPSS as a pair. Tensr’s item is Time series → Decomposition → Stationarity Tests. The path string stored for this item says Time series → Diagnostics → Stationarity. The item itself sits under Decomposition with STL.

ARIMA / SARIMA is the forecast that usually follows. Autocorrelation shows leftover lag structure.