Summarize measurements

Use Statistics to summarize object measurements at the object, image or declared-sample level, while keeping counts, exclusions and units visible.

New in 0.16.0a1

Statistics expands Summarize Measurements. New nodes use the descriptive recipe below. Older saved summaries retain their previous calculations until you explicitly upgrade them. No inferential tests, p-values, significance labels or confidence intervals are included.

Start with a measurement table

1. Connect a calculated measurement table to Statistics, under Measurements → Tables. A single image is enough; a batch collection is optional.
2. Review Auto-select measurements, or switch it off and choose the Measurements you need. Under Group results by, leave None for one overall summary or choose a field such as treatment.
3. Set Each observation represents using the table below. For image/sample averages, select the relevant Image identity column and Sample identity column. Under Statistics to report, choose the reductions.
4. Set Missing or invalid measurements to Exclude and report or Stop and review. Never interpret an excluded measurement as zero.
5. Calculate and inspect the result table. Check the counts and status fields alongside the values before exporting or plotting.

Choose Open Results Workspace… to keep these controls beside the updating summary table, with Data and Plots in the same window. It edits this same Statistics node, not a separate recipe. See Explore results in one workspace.

The initial selection is count, mean and sample SD over objects. You can also choose median, lower/upper quartiles, IQR, minimum, maximum and sum. The Statistics reference defines each method and explains empty groups and undefined results.

Hover over a field or choice for its explanation. Auto-select measurements tracks eligible numeric measurement columns as the upstream table changes. It excludes recognized identities/metadata and the chosen grouping and identity fields. It neither creates measurements nor converts text to numbers.

New in 0.16.0a4: paired measurement selection

Use Select all above Measurements to keep the currently eligible numeric measurement names as a fixed manual selection, or Deselect all to clear all selected measurements and choose a smaller set. Select all skips IDs, text, grouping/identity fields and other fields marked not automatic. Both actions turn off Auto-select measurements.

An empty selection cannot be calculated: choose at least one measurement first. These actions affect the analysis, unlike the Results Workspace Data tab's visible-column controls. They work identically in the inspector and Results Workspace.

The controls are divided into Measurements, Grouping, Observation unit and Statistics to report. The short note below the missing-value choice explains the selected policy; hover for the full details. Sample SD still needs at least two valid observations at the chosen level.

Choose groups

Group results by calculates a separate summary for each distinct value of one column. For example, choosing condition separates control and treatment. None combines all eligible rows into one summary.

Check Group by several columns to reveal the grouping checklist. Select, for example, condition and timepoint to summarize each distinct combination: control at time 0, control at time 1, treatment at time 0, and so on. Merely opening the checklist keeps the current grouping until you change a selection. Unchecking Group by several columns returns to None (no grouping), not the previous single-column choice.

Choose what is being summarized

Level	What enters the final summary	When unequal counts matter
Objects	Every eligible measurement row	An image with more objects contributes more values.
Image averages	One arithmetic mean per contributing image	Each image mean has equal weight, regardless of object count.
Sample averages, Equal images	One mean of image means per sample	Images have equal weight within each sample.
Sample averages, Equal objects	One pooled object mean per sample	Images with more eligible objects have more weight within that sample.

The sample setting Within each sample, weight selects Equal images or Equal objects. Both modes give each contributing sample mean equal weight in the final summary. For example, one image with measurements 2, 4 and another with measurement 12 have an equal-image mean of 7.5, but a pooled-object mean of 6. Neither weighting is universally correct: choose the quantity you intend to describe and keep that choice with the result.

Choose image and sample identities

• Image identity column: all object rows from the same image must have the same ID, and different images must have different IDs across the collection. For example, every object from field image_01 shares that image ID. Object labels such as label_id identify objects, not images.
• Sample identity column: use the same ID for images from the same declared experimental sample, such as an animal or culture. For example, image_01 and image_02 can both belong to sample animal_01. Choose this mapping from your experimental design; VIPP does not infer it from filenames or prove that samples are independent.

An identity marked optional counts only adds image/sample counts; it does not change the averaging level or weighting. Switching the observation level clears identity choices that are no longer used; review the visible fields after switching.

Image averages require one group per image; sample averages require one group per sample (and, with equal-image weighting, one group per image). Objects may belong to different categories within one image. Conflicting image-to-sample assignments are always errors, not silently split populations.

An image with no eligible measurement contributes no mean and no invented zero. For a collected batch, review empty, failed and excluded images in the collection review, not just in the summary.

Keep and use the result

Save the workflow to keep selected columns, level, weighting and exclusion policy. The output is an ordinary table: each measurement has separate statistic and count columns, with its own units. Method fields and a saved recipe are included so a result is not detached from its averaging rule.

The inspector keeps the short summary separate from the full evidence:

• Result overview reports the number of summary rows and a compact inclusion breakdown for each measurement, totalled across all groups. Compare objects used/excluded with the number of observations summarized: at image/sample level, those observations are averages, not individual objects. The full table keeps the per-group details.
• The compact Results preview shows group values, valid n and selected statistics.
• Open in window and Export CSV/TSV… include the complete table, with exclusion reasons, image/sample counts where supplied, and the saved recipe.

Connect the output to Plot Results to plot selected result columns; remember that these rows already represent summaries, not the original objects. The separate batch collector's Excel export does not make ordinary Statistics table export an Excel exporter.

Existing summaries

Opening an older workflow keeps its legacy summary behavior, including a sample SD of zero for a singleton. Review the legacy notice and choose Upgrade to descriptive Statistics only when you intend to adopt the new rules. The upgrade can be undone. Version 2 reports singleton sample SD as undefined, distinguishes exclusion reasons and uses the selected aggregation level. Save a copy of a consequential workflow before changing its analysis recipe.

These are descriptive results. Object counts, image counts and declared sample counts are not interchangeable evidence of biological replication.