How long statistical analysis takes for a manuscript, and what makes it slow

The three stages, and where time actually goes

Data preparation covers everything between the raw export and a dataset that can be analyzed: mapping columns to items, recoding text responses, reverse-coding, scoring instruments, handling missing items, merging waves or sites, and writing down each decision. For survey data this is the largest stage, and the one most often underestimated in project timelines.

Analysis is fitting the pre-specified models, producing tables and figures, and checking assumptions. When the plan is clear and the data are prepared, this stage is measured in days rather than weeks. Reporting is the drafting of methods and results text and the iterations with co-authors, which restarts whenever the analysis changes.

What stretches preparation

Exports that store response labels instead of codes. Items that were added, dropped, or reworded between waves. Instruments built into the survey without their scoring manual. Free-text fields that hold information the analysis needs. Multiple exports that have to be merged on an identifier that is not quite consistent. Each of these is solvable, and each adds a round of questions back to the study team, which is where calendar time disappears.

In the RRIPG injury questionnaire project described on this site, the data preparation stage went from two weeks to three days once the structure was documented and the checks were systematic. That is one project's number, not a benchmark or a general turnaround estimate. In that project the gain came from preparation, not from faster statistics.

What stretches analysis

An analysis plan that is not written down. Every unplanned question becomes a new model, a new table, and a new decision about how to label it. Changing the outcome definition after seeing results. Missing data that need a method chosen and justified rather than a default applied. And a primary comparison that turns out to need a different model than assumed, for example a clustered design that was treated as independent.

What stretches reporting

Methods text written from memory instead of from the record of decisions. Tables produced by hand, so every change to the analysis means rebuilding them. Co-author rounds that reopen analytical choices late. Keeping the analysis code, the decision log, and the tables connected is what makes a revision an afternoon instead of a fortnight.

What a research team can do before handing over data

Most of the delay lives in the handover. Teams that arrive with the following shorten every stage that follows.

• The research question and the primary comparison in one sentence each.
• The instruments used, with versions and scoring manuals or references.
• The export with a data dictionary, even a rough one: column, item, response options.
• A note on anything that changed during collection: items, waves, sites, platforms.
• The deadline that matters and who needs to sign off on the results.