SPSS Studio

SPSS Studio is an exceptionally powerful integrated development, execution, and interpretation environment for IBM SPSS Statistics. It grew out of a vision that its designer, Jing LIN, has pursued ever since he began teaching statistics courses at university: to integrate the software's professional syntax highlighting and completion, computational-engine connectivity and execution, dataset management and analysis, and AI-assisted question answering into a unified workflow, enabling professional researchers to conduct statistical analyses efficiently, conveniently, and elegantly. Its designer believes that SPSS Studio will infuse this venerable, nearly sixty-year-old statistical package with a youthful spirit.

SPSS Studio 是一个功能极端强大的IBM SPSS Statistics集成式开发、执行与解释环境。 它源于设计者Jing LIN在大学讲授统计学课程以来念兹在兹的愿景：将该软件的专业语法高亮与补全、计算引擎链接与执行、数据集管理与分析以及 AI 问答交互，和谐地整合在统一的工作流中，助力专业研究人员以高效、便捷、优雅的方式执行统计分析。设计者相信，SPSS Studio将赋予这个拥有近60年历史的古老统计软件以年轻的气质。

Author：Jing LIN (林景)

Affiliation：Associate Professor, Nanjing University of Finance and Economics

Contact：linjing@nufe.edu.cn

SPSS Studio is independently developed. It is not affiliated with, endorsed by, or sponsored by IBM.

Features

• Generated TextMate highlighting based on a 309-command IBM SPSS Statistics syntax inventory, with distinct command, subcommand, variable, function, format, literal, operator, macro, and comment families.
• Optional SPSS Studio Light and SPSS Studio Dark color themes for a complete, contrast-checked syntax palette; installation never changes the user's active theme.
• One-character local completion with ranked commands, minimal templates, gray input hints, and optional AI suggestions for the next SPSS command.
• Editor-title actions for Undo, Run, and Run All.
• A standard File → New File… → SPSS File entry that opens an empty SPSS editor beside Studio without running syntax.
• A persistent, serialized local SPSS backend.
• Native statistical output in Simplified Chinese for Chinese VS Code display languages, and in English for English or other display languages.
• Native OMS HTML output with run history, structurally aligned Markdown tables for concise AI interpretation, safe standalone HTML export, and browser-based printing.
• Read-only Data and Variables views for the current Active Dataset.
• A read-only Variables table with Name/Label filtering, persistent multi-selection, Copy, Insert, and bounded AI-assisted Explore actions; Name cells can also be double-clicked for immediate insertion.
• A lightweight SPSS AI question-and-answer workspace embedded in the fourth SPSS Studio tab, Chat, with multiple DeepSeek, Zhipu GLM, Qwen, Doubao, and custom OpenAI-compatible model profiles.
• Copy and Insert controls on every AI response code block, plus an explicit Run action on SPSS blocks; generated code is never run automatically.
• A bilingual, locally bundled Quick Start page that opens after installation and—unless the user opts out—once after each future update, with a permanent book action in every .sps editor.
• Automatic IBM SPSS Statistics discovery on macOS and Windows, with manual path overrides.

Requirements

• Visual Studio Code 1.95 or newer.
• A locally installed and licensed IBM SPSS Statistics with the Python Integration Package.
• A trusted VS Code workspace for execution and Active Dataset access.

IBM SPSS Statistics 27 has been tested on macOS. SPSS Studio includes bridge syntax and UTF-8 transport compatibility for the Python 3.4 runtime supplied with SPSS 26, native output-language control, compatible Windows CRLF syntax submission, and local completion. Windows-specific command tests are included but skipped on macOS; execution against a real Windows SPSS installation, including SPSS 26, remains pending user acceptance. A passed mock or grammar test is not a real Windows acceptance result.

No separate Python installation is normally required. On Windows, SPSS Studio requires IBM's statisticspython3.bat launcher so that the Python runtime and SPSS installation remain version-matched; it does not silently fall back to a generic Python executable. On macOS, IBM's bundled Python remains available as a compatibility fallback when the IBM launcher is absent.

Installation

Install SPSS Studio from the Extensions view in Visual Studio Code, then open a .sps file. VS Code automatically assigns the SPSS Syntax language mode.

To start without an existing file, choose File → New File… → SPSS File (文件 → 新建文件… → SPSS 文件). The extension opens a fresh, unsaved document in SPSS language mode on the left and Studio on the right. Save it as .sps when ready. This uses VS Code's standard extension menu API, requires no editor modifications, and does not start the engine or execute syntax.

On first activation after installation, SPSS Studio opens Quick Start for an English VS Code interface or 快速上手 for a Chinese interface. The page explains the complete editor, execution, Studio, Variables, and Chat workflow. Its prominent update-reminder checkbox controls whether later extension versions open the page once after updating. This preference stays in local VS Code global state and is not written beside an .sps file or packaged in the VSIX. The page remains available from the book icon in every .sps editor or from Settings → System in Studio.

On Windows, use Detect and Configure SPSS in Quick Start; the progress window opens before detection begins. You can also run SPSS: Detect and Configure SPSS from the Command Palette at any time. If automatic discovery fails, choose the SPSS installation, its bin directory, or its IBM/SPSS parent folder. Setup distinguishes missing IBM Python integration from residual installation files and records each attempted installation separately. WSL, SSH, and container windows display a local-window requirement; they cannot connect to the SPSS installation on your Windows desktop. Local macOS Quick Start remains unchanged.

Detection covers SPSS on non-system drives using an independent, time-limited disk scanner, alongside registry and shortcut hints. PowerShell and WMI are not prerequisites for this disk scan; system tools are resolved from Windows directories without requiring PATH or ComSpec. Automatic scanning is bounded to SPSS and common software folders. For arbitrary deep/custom locations, select the installation folder directly. Reports distinguish OS/process architecture and ARM hints. IBM does not officially support SPSS clients on Windows ARM; a setup pass does not constitute IBM compatibility certification. HarmonyOS is a separate platform, and a Windows VM must run VS Code and SPSS together inside that Windows environment.

Running syntax

The engine uses the VS Code display language for native table headings: Chinese interfaces request Simplified Chinese; English and other interfaces request English. This is independent of Chat's response-language setting. SPSS supplies its own translations, and existing outputs retain the language used when they were generated. The corresponding SPSS language resources must be installed; inability to apply the requested language is reported explicitly.

GET FILE completion recommends GET FILE "path/to/data.sav". without its optional equals sign. String-bearing templates and Chat suggestions use ASCII double quotes; other required equals signs and valid handwritten syntax remain supported. Windows CRLF and Unix/macOS LF source files use the same submission path without changing the source file on disk.

Current-command execution preserves blank lines inside a command until its terminating period, matching SPSS's Python Submit backend. Blank lines between complete commands remain intact. Use ASCII periods to terminate commands; blank lines do not repair a missing period in this execution mode.

Local completion suggests commands, variables, and functions after one character. Typing / offers subcommands for the current procedure. Tab accepts the selected candidate. For example, typing G and pressing Tab inserts GET FILE "". with the cursor between the quotes. Gray field hints follow the VS Code interface language. Templates show each SPSS subcommand on its own line. Fill the last field and press Enter to move past the template's closing punctuation and start a new line. Ctrl+Space opens the full local list.

For optional next-command suggestions, select Coding on a model in the Studio Settings tab. Enabling it scans the active SPS file locally. After a complete command and a new line, Studio either offers a local suggestion or sends the selected model an anonymized structural summary of the active file. Tab inserts the gray suggestion for review. Run it separately when ready.

• macOS: Cmd+Enter
• Windows/Linux: Ctrl+Enter
• Command Palette: Run

With a selection, Run submits the selected syntax. Without a selection, it identifies and submits the current SPSS command or structural block. Run All submits the complete in-memory document, including unsaved changes. Undo uses VS Code's native undo stack.

The first execution starts one persistent local SPSS processor. Run, metadata, and data-page requests share one serialized queue so that two calls never enter SPSS concurrently.

SPSS Studio expands leading TAB indentation on SPSS command lines only at the final submission boundary. This avoids the SPSS 27 external Python Submit failure caused by literal TAB characters while preserving quoted strings and BEGIN DATA, BEGIN PROGRAM, BEGIN GPL, BEGIN SCRIPT, and MATRIX bodies. The source document is never modified.

Output

Each execution uses one private, uniquely tagged, short-lived OMS HTML destination. The selected result initially receives the full Output width. Use History beside Print to show or hide the resizable Runs column on the right.

• Explain uses the left-hand checkboxes to explain individual tables or result blocks, including separate tables from one analysis. Multiple checked items are sent in one request for separate numbered interpretations; with none checked, the whole run is explained. Native HTML row spans, column spans, layered headers, captions, and footnotes are converted to aligned Markdown. Safe model qualifications are retained; unchecked table bodies, figures, Notes, command echoes, paths, and run metadata are excluded.
• Export saves the selected output as a sanitized, self-contained HTML file.
• Print opens a print-ready copy in the default browser.
• SPSS: Clear Output removes output files from the current extension session without changing the Active Dataset.

The extension displays native SPSS HTML rather than reconstructing statistical tables. It does not generate SPV or OXML files and does not use SpssClient.

The Output action bar remains visible while results scroll. Export and Print use the same checked tables/text blocks as Explain; with none checked, they retain the complete output, including figures. Partial export/print preserves the selected native HTML, styles, titles, and footnotes without AI row/character limits. Invalid selections stop the action. Figures currently have no individual checkbox. Creating a print-ready file opens the browser's print dialog; it does not silently submit a print job.

Execution statuses are SUCCESS, SUCCESS_NO_OUTPUT, WARNING, ERROR, ENGINE_ERROR, and TIMEOUT.

Data and Variables

SPSS: Show Data opens a read-only view of the current Active Dataset. Rows are paged, while variables are available through continuous horizontal scrolling. The bridge never calls fetchall and limits each request to 500 rows and 200 variables.

SPSS: Show Variables displays variable order, Name, Label, Type, Format, and Measure. It reuses metadata already cached for completion and Data preview and does not modify the SPSS data dictionary.

• Select variables with the checkbox column. Copy writes their names to the clipboard in dataset order, separated by spaces. Insert places the same list at every current .sps cursor or selection as one undoable editor operation.
• Use the bottom-right Filter field to match either Name or Label by a case-insensitive substring, including Chinese text. Filtering changes only the visible rows: selections outside the current result remain selected, and Copy, Insert, and Explore continue to use all selected variables.
• Double-click a cell in the Name column to insert that exact cached variable at the most recently used .sps selection. Other cells remain read-only.
• Explore sends the selected variables to the current Chat; with no selection, it uses the first 10 variables in the current filtered result. One request accepts at most 20 variables. The local SPSS bridge sends no raw case rows: it builds bounded profiles containing dictionary metadata, up to 100 value labels, up to 20 observed categories with frequencies, or compact continuous/date summaries. Very high-cardinality categorical counts are explicitly approximate.

Explore is bounded, but it is not zero-disclosure: observed categorical or string values can themselves contain sensitive text. Review the variables before using Explore and do not send sensitive values to a provider that is not authorized to receive them.

Variable Explore is measurement-aware. For one variable, Chat proposes descriptive syntax and next steps. For two or three variables, it proposes only statistically appropriate crosstabs, plots, correlations, or regression syntax and states its assumptions. For more than three variables, it first proposes possible themes and asks the user to clarify the intended analysis. Explore replies always end with a short reminder that these are suggestions rather than formal results.

SPSS AI

Under Settings → System, reopen Quick Start, switch the VS Code theme, set separate font sizes for the SPS editor and Studio panel, or change the AI chat language. Font sizes can be typed or picked from the list, and Default restores the inherited size.

Select text in the SPS editor and check items in Studio before using Explain in Output or Explore in Variables. Each button sends the selected code with its checked items. Explain in Chat from the editor can send the selected code with both checked sources in one request. A failed run provides a selectable diagnostic even without native HTML. A cursor without a text selection retains the existing single-source actions.

The reusable SPSS Studio panel opens beside the native .sps editor and contains Variables, Data, Output, Chat, and Settings in that order. Opening a .sps file alone does not open the panel. Running syntax opens Output; the Show Data, Show Variables, Show AI Assistant, and Manage AI Model Profiles commands open their corresponding Studio location. The extension does not contribute or automatically reveal a native VS Code bottom-panel view. Editor menus, Studio tabs, controls, progress states, tooltips, and direct feedback follow the VS Code display language: Chinese locales use the Simplified Chinese catalog. All other locales use English. This interface language is independent of the model response-language preference.

The Chat tab is an SPSS AI question-and-answer tool. Its compact English header places Active model on the left, followed by New, Current, and History. A Simplified Chinese VS Code interface displays the corresponding concise Chinese controls. The boundary between the transcript and question box can be dragged with the mouse or adjusted from the keyboard. Completed answers render common Markdown structures without accepting model-generated HTML. A concise built-in system instruction grounds replies in executable SPSS Syntax, applied social statistics, assumptions, key options, and result interpretation without claiming that generated syntax was run. Replies default to Simplified Chinese and stay concise unless the user explicitly asks for detail.

Chat is deliberately scoped to SPSS Syntax, IBM SPSS Statistics use, statistical methods implemented in SPSS, and SPSS output interpretation. Clearly unrelated requests receive a short boundary notice rather than a general-purpose answer. This instruction-level boundary reduces off-topic responses but is not a security guarantee against every possible model behavior.

In an .sps editor, right-click a selection and choose Explain in Chat. With no selection, the command scanner sends the complete SPSS command or structural block at the cursor. The Studio panel opens to Chat and submits a concise explanation request immediately; it never executes the selected syntax. If no usable model profile exists, the command opens Settings instead.

Open the Studio Settings tab beside Chat, or run SPSS: Manage AI Model Profiles.

1. Create a profile and choose DeepSeek, Zhipu GLM, Qwen, Doubao, or Custom OpenAI-compatible.
2. Confirm the prefilled Base URL.
3. Enter the model identifier supplied by the provider.
4. Enter an API Key and save. A blank profile name is filled automatically from the provider and model.

Each built-in provider profile also has a Reasoning checkbox, which is off by default. DeepSeek, Zhipu GLM, and Doubao receive their native thinking.type field; Qwen receives enable_thinking. Custom OpenAI-compatible profiles keep the checkbox disabled because the extension does not guess provider-specific fields. Whether a particular model supports the selected mode still depends on that provider and model.

Settings lists profiles as Model 1, Model 2, and so on. Select a number to see and edit its details. Test checks the current connection using a fixed short message. Activate chooses the model for Chat. Coding can use a different profile. The API Key field stays blank for saved profiles and displays the saved status below. A changed provider or Base URL requires a new key.

Language in Settings is shared by ordinary Chat questions, Explain in Chat, and Output Explain. It defaults to Chinese (default) and can be changed to English. The preference is stored in this extension's global VS Code state rather than in an .sps file or workspace folder.

Provider model catalogs change independently of the extension, so the model field is intentionally editable. The four built-in presets use these official OpenAI-compatible Base URLs:

Every fenced code block in an assistant response has high-contrast actions ordered as Copy and Insert. Blocks identified as SPSS add Run as the final action and use theme-aware highlighting generated from the same SPSS language schema as the editor TextMate Grammar:

• Copy writes the same code to the system clipboard.
• Insert places the code, without fence markers or the language identifier, at the most recently used .sps selection.
• Run submits the complete SPSS code block directly to the existing serialized local execution queue without inserting or changing the .sps document, then switches Studio to Output as execution starts.

Insert performs one ordinary, undoable text edit. It does not save or execute the syntax. Run remains a deliberate user action, requires a trusted workspace, and never runs automatically when a response arrives.

The API Key is stored in VS Code SecretStorage, not in settings.json, logs, chat history, the repository, or the VSIX. Each request contains a fixed SPSS-assistant system instruction, bounded text from the active conversation, and content the user explicitly submits. Explain in Chat sends the exact selection or resolved current command. With an explicit selection, it can also include checked Output results and Variables summaries. Output Explain sends a local Markdown conversion capped at 50 rows per table and 30,000 characters in total; it removes Notes tables, runtime/provenance metadata, command echoes, file paths, scripts, styles, and every figure before the request is built. Variable Explore is the only dataset-aware AI action: its bounded profiles contain no raw case rows and are capped at 30,000 characters before any selected syntax or output is attached. Ordinary Chat requests do not attach the rest of the SPS file, variables, cases, Output, filenames, or workspace paths.

Up to 100 conversations are saved as plain text in this extension's private local global-storage directory. The same history is available for all .sps files and workspaces in the same local VS Code profile; no file or folder is created beside an .sps document. History is not placed in Settings Sync and is not packaged in the VSIX. New starts a blank conversation without deleting older history. Individual conversations can be opened, renamed, or deleted; clearing all history requires two confirmations.

Model output is untrusted and can be wrong. Inspect generated syntax before running it. SPSS Studio does not provide model-provider billing, retention, or correctness guarantees.

Commands

• SPSS File
• Run
• Run All
• SPSS: Undo Last Edit
• SPSS: Start Engine
• SPSS: Stop Engine
• SPSS: Restart Engine
• SPSS: Show Status
• SPSS: Show Output
• SPSS: Show Data
• SPSS: Show Variables
• AI Assistant
• SPSS: Manage AI Model Profiles
• Explain in Chat
• SPSS: Refresh Data Preview
• SPSS: Clear Output

Configuration

Built-in examples and session data

Quick Start includes Load Example (or 一键加载示例 in a Chinese VS Code interface). It opens an English or Chinese .sps tutorial together with an already packaged, original .sav dataset: 60 fictional cases and id age gender edu income hours. No download or data generation is needed. The tutorial demonstrates import, labels, codebook, descriptive statistics, bar/scatter charts, an independent-samples t test, regression, and Studio/Chat interactions. Opening it does not run syntax or contact an AI provider. Editable copies stay in the extension's private example directory, and reopening preserves edits.

The Output toolbar's Clear, between Print and History, discards all loaded datasets and unsaved changes in this Studio engine after confirmation. It preserves source files, output history, and Chat. The Runs sidebar's separate Clear still clears output history only.

Data-loading commands can succeed without producing statistical tables; their run now includes the available case and variable counts. SPSS Warning #67 indicates a file-use condition and does not alone establish a failed read. Its native text is retained with concise guidance. Studio preserves named datasets rather than silently clearing them before imports; use Clear when you intentionally want an empty session. Another process's file use can still require action outside this session.

Runtime configuration

On Windows, select Detect and Configure SPSS in Quick Start, or run SPSS: Detect and Configure from the Command Palette. The wizard finds installations, runs an isolated temporary analysis, saves only the necessary SPSS Studio settings, and verifies the effective saved configuration. It checks three rows, the expected mean of 30, native HTML output, and graceful shutdown. No system-wide Python variables need to be edited.

An active SPSS session is preserved: save any data you need and stop the engine manually before running setup. If multiple installations pass, select one. If none are found, use Choose SPSS Folder. Conflicting workspace values require confirmation for targeted repair. Failed or cancelled configuration restores the previous values; concurrent user edits are preserved and reported as conflicts. Interrupted setup changes are journaled for recovery after reload.

On Windows the result is saved as a concise desktop TXT report when the desktop is writable; Save Desktop Report offers a save-location fallback. Detailed bounded setup diagnostics stay in the extension's private global-storage directory. Setup never sends data to Chat. On macOS the same command is available manually, but the Windows Quick Start action and automatic desktop report are not enabled.

On Windows, JSON settings paths must escape backslashes, for example C:\\Program Files\\IBM\\SPSS Statistics\\32.

Workspace trust and local processing

Syntax highlighting and static completion remain available in an untrusted workspace. Starting SPSS, executing syntax, accessing the Active Dataset, and sending AI network requests require a trusted workspace.

SPSS syntax and dataset requests are sent to the locally installed IBM SPSS Statistics processor. The extension does not provide a cloud execution service.

AI questions are sent directly from the extension host to the Base URL selected by the user. HTTPS is required, except that loopback HTTP addresses such as localhost and 127.0.0.1 are permitted for local compatible servers. Redirects are rejected so the bearer credential is not forwarded to another origin.

Development

Node.js 22 or newer and npm are required.

npm ci
npm run verify
npm run package

npm run verify regenerates and checks the grammar, compiles TypeScript, runs ESLint, executes the TypeScript unit tests, and executes the Python Bridge tests. npm run package repeats verification before creating the VSIX under dist/.

Architecture, syntax coverage, and troubleshooting details are available in:

• Architecture
• Syntax coverage
• Troubleshooting

Report reproducible defects through GitHub Issues.

Limitations

SPSS Studio is not an LSP or a full semantic validator. It does not provide statistical menus, editable Data View cells, variable metadata editing, paper-table generation, or a general SPV viewer. Variables is intentionally read-only, and exported output is HTML rather than SPV. The AI client supports the common OpenAI-compatible Chat Completions contract for streaming chat and short completion requests. Provider-specific tools, search, attachments, and advanced parameters are outside its scope.

Version History

Detailed release notes are available in CHANGELOG.md.

License

SPSS Studio is proprietary software distributed free of charge for permitted use. The source code is publicly visible for review but is not open source. Modification, derivative works, sale, sublicensing, republication, redistribution, and repackaging are prohibited unless prior written permission is granted. See LICENSE for the complete terms.

Copyright © 2026 Jing LIN (林景). All rights reserved.

IBM and SPSS are trademarks of International Business Machines Corporation, registered in many jurisdictions worldwide. All other trademarks belong to their respective owners.