DID-ACTO / FIELD GUIDE
Prepare and upload data
A clean file is only one ingredient. The structure and research definitions matter too.
Choose a supported format
CSV is a straightforward starting point. The uploader also accepts TSV and related delimited text, Excel .xlsx/.xlsm, Parquet .parquet/.pq, and supported JSON/NDJSON table representations. Stata .dta files are supported for format releases 114, 117, 118 and 119. For long strL strings, save fixed-width strings or export to CSV or Parquet. Export legacy Excel .xls to .xlsx or CSV first.
For Excel, use a single table with one header row. For JSON, use a table of records with consistent fields. Inspect the uploaded dataset to confirm the columns and types.
Name and describe your dataset
Choose a file, then set its filename and an optional description before uploading. Include the source, period, unit of observation and intended project. Keep the file extension unchanged. Descriptions support up to 4,000 characters.
Use Edit details in Files and reports to update either field later. Search the library by name, description or upload identifier. Copy dataset context includes your description and stable reference for your assistant; review the description with your assistant when confirming the study design.
Renaming changes the library and download name. The original file bytes, upload identifier and existing analysis references stay intact. Metadata is private to the workspace and its administrators and is included in workspace backups. Older uploads can be labeled manually. Conflicting edits ask you to refresh before saving again.
Understand a panel
A panel follows the same units across time. A common layout has one row per unit and period, a consistently measured outcome, and enough information to identify treatment and its timing.
| Column example | What it means | Check before analysis |
|---|---|---|
unit_id | A stable unit such as a store or region | Do identifiers mean the same unit in every period? |
period | An ordered time index | Is the frequency consistent and ordering unambiguous? |
outcome | The quantity you want to study | Are units and transformations consistent? |
treated | Treatment status for this observation | Does its coding agree with the rollout? |
| Covariates, if used | Pre-treatment adjustment variables | Could treatment itself have changed them? |
These names are examples, not mandatory column names. Explain the actual schema to your assistant and confirm its interpretation. Do not silently replace missing outcomes with zero or fabricate dates for never-treated units.
See the shape
| unit_id | period | outcome | treated |
|---|---|---|---|
| store_a | 1 | 100 | 0 |
| store_a | 2 | 108 | 0 |
| store_a | 3 | 120 | 1 |
| store_b | 1 | 95 | 0 |
| store_b | 2 | 101 | 0 |
| store_b | 3 | 105 | 0 |
A tiny format illustration, not a dataset with sufficient support for inference. Use a shared sample for a complete practice analysis.
Upload into your workspace
- Sign in to the portal with your own account.
- Choose the file under Upload a dataset.
- Select Upload or resume and wait for completion.
- Copy the upload identifier or MCP reference returned by the portal.
- Give that reference to your assistant and ask it to register and inspect the dataset.
Once uploaded, register the file with your assistant, inspect its structure and review the analysis plan.
Use references, not laptop paths
An upload identifier looks like upl_…. Use the actual value from your workspace; the example is not a usable identifier. A shared sample or an administrator-imported private dataset may instead have a server path shown by the portal.
Your server cannot read C:\Users\… from your laptop. Do not paste local paths or raw rows as a substitute for uploading the file. Keep any private dataset references within your authorized study context.
Respect both file and analysis limits
Each upload is limited to 200,000,000 bytes. Your 2,000,000,000-byte workspace also contains runs, reports and caches; filesystem overhead reduces usable capacity. Additional parsing, row, cell and expanded-memory limits may reject a file below 200 MB.
A small compressed or columnar file can expand substantially in memory. Start with the necessary columns and a documented study sample. See troubleshooting if upload or inspection fails.