Switching from JMP: A Field Guide

Whether to leave JMP is covered in Stratum vs JMP. In short: JMP costs $1,390 per user per year on its store, JMP Pro costs $8,820, and both are sold only to organizations. Stratum is a native Mac & Windows app with a one-time price of $119 or $249. This post is about the move itself. Your JMP habits are good ones, and most of them carry over, sometimes under the same name.

The concept map

Modeling types: the habit that carries over intact

JMP taught you that a column's modeling type matters more than its storage type, and Stratum works the same way. Every column gets a measurement level from the moment the file opens, and Stratum reads it from the data. A fully value-labelled numeric column is a set of codes. A column whose levels you have put in order is ordinal. A small set of whole numbers that is a minority of the values present is nominal. The level decides which columns are offered as measurements and which as groupings. If Stratum reads a column wrong, open Column Definition… and set it, with Apply to covering as many columns as you tick. The forty Likert items in a survey can be declared in one pass.

Fit Y by X becomes Guide Me

Guide Me (in the toolbar, and at the top of the + menu) asks what you want to find out, then which columns, with its best guess already filled in. The routing will feel familiar. It follows Fit Y by X's table on the two measurement levels:

• Continuous by continuous leads to Correlation, shown as a scatter.
• A measurement by a group leads to Two-Sample or One-Way ANOVA.
• Two categorical columns lead to a Contingency Table.

The panel also prints what it read, for example Reading Conductivity (Continuous) by Shift (Nominal). The one cell Stratum can't fill is logistic regression. When the response is categorical and the factor is continuous, Guide Me compares the measurement across the groups instead and says that it swapped them. See Which statistical test? for the whole decision table.

Row states, rethought

JMP packs selection, exclusion, hiding, labels, colors and markers into one row-state bit field. Stratum splits them apart on purpose:

• Marks are the persistent "these rows matter" flag. Select rows in the data sheet's row gutter and choose Data ▸ Rows ▸ Mark Rows, or shift-click points on a chart. Every chart then rings those points. You can filter on is marked and use marked() in formulas.
• Colors and markers belong to the row, either set directly or live by column, and every point chart draws them. Partition vs row colors explains when to use each.
• Label text is data. There is no label column role. Each chart's Labels control picks the columns to name points from, so two charts can label the same rows differently. Labels can also be dragged into place. Label points on a chart walks through it.
• Exclusion is a filter, so it is visible, reversible and saved with the file. There is no Hide row state.

Validation columns, with or without a column

In JMP Pro, honest model assessment starts with a validation column. In Stratum, Decision Tree, Random Forest, Boosted Trees and Linear Regression each have Validation and Test Set sliders, stratified by response by default. Models given the same fractions hold out the same rows, so Model Comparison can rank them on rows none of them saw. To fix a split for good, Save as Column writes it as a labelled 0/1/2 column. To keep the split you already made in JMP, choose it as the Validation Column. Stratum reads 0, 1 and 2 or the words Training, Validation and Test, so your JMP column works as it is.

Moving your data across

Stratum doesn't open .jmp files. From JMP, save the table as Excel or CSV and open it in Stratum. Stratum also reads SPSS .sav and Stata .dta directly, with value labels and missing codes intact. Here's what to expect:

• Formulas arrive as values. Rebuild the ones you still need with Add Column….
• Column properties such as value labels and value order don't travel in a CSV. Re-declare them in Column Definition, one pass for many columns.
• Modeling types are re-inferred, so check the level shown in the column rail.
• On a wide file, the column rail is the index. Navigate a wide dataset shows how.

What you'll leave behind

Stratum has no JSL and no design of experiments, reliability and survival analysis, or mixed models, and it doesn't aim to replace them. Logistic regression is not yet there. Classification trees and forests cover categorical responses for now. Graph Builder's drag-and-drop canvas is also absent: Stratum charts are built from control-bar pickers. That trades JMP's improvisation for a chart that states its settings and reopens exactly as you left it.

Download Stratum Stratum vs JMP →

Frequently asked questions

Can Stratum open JMP (.jmp) files?

No. Save the data table from JMP as Excel (.xlsx) or CSV and open that in Stratum. Stratum also reads SPSS .sav and Stata .dta files directly, with their value labels and missing codes. Formulas arrive as values, so rebuild the ones you need as Stratum formula columns.

Does Stratum have JMP's modeling types?

Yes, under the same three names: Continuous, Nominal and Ordinal. Stratum calls it the measurement level. It is read off the data when the file opens, and you can override it for any number of columns at once in Column Definition.

What replaces row states in Stratum?

Rows can be marked, colored and given markers from Data ▸ Rows or a point's right-click menu, and every chart honors them. Exclusion is done with the filter in the Organize panel, or with Exclude Point from Limits on a control chart. Label text always comes from columns. There is no Hide row state.

Will my JMP validation column work in Stratum?

Yes. Decision Tree, Random Forest, Boosted Trees and Linear Regression each take a Validation Column coded 0, 1 and 2, or Training, Validation and Test, so a JMP validation column exported with the table works as it is. You can also hold rows out by fraction and freeze the split with Save as Column.

Is there a Fit Y by X in Stratum?

Guide Me routes on the measurement levels of the two columns you pick, much as Fit Y by X does: two continuous columns lead to Correlation, a measurement by a group leads to Two-Sample or ANOVA, and two categorical columns lead to a Contingency Table. The one cell without a Stratum analysis is logistic regression.

What will I miss from JMP?

JSL scripting, design of experiments, reliability and survival analysis, mixed models, and logistic regression, which Stratum does not have yet — logistic and Poisson arrive in 1.6. Graph Builder’s drag-and-drop canvas is also absent; Stratum builds charts from control-bar pickers, so a chart states its settings and reopens exactly as you left it.