Analysis in the chat

How Iris computes over every answer, where her numbers come from, how she quotes answers and sorts open answers into a theme list.

Iris also analyzes your surveys in the chat. Ask, for example, "How does purchase intent differ between full-time and part-time employees?" or "Compare survey runs 1 and 2 on the price question". Iris computes over every answer of one or more survey runs and single questions to audiences. That includes crosstabs of two questions, splits by two attributes and comparisons across several survey runs.

Calculations

Iris can explore data and calculations without saving a calculation every time. Result numbers she uses in the chat, a report or a presentation come from a saved Calculation. Iris shows the results directly in her answer as numbers or tables. The Base is the number of people behind a value.

The status line shows whether the calculation is running, completed or failed.

A saved calculation never changes. To find out how a number came about, ask Iris. She explains the computation in words.

For large calculations, Iris can read further rows and the output produced during computation, for example when you ask “Check the last groups too”. She does not need to run the calculation again. This also works in later messages while the chat and access to every source remain available. Older output without a saved copy remains unknown; Iris does not describe it as empty.

What Iris computes in the chat stays in the chat. The report shows only its survey run's computed results. To keep a result, including a comparison of several survey runs or numbers from single questions and files, put it into a presentation.

You can also ask Iris to check a report statement against the original answers. She can read both and examine the supporting evidence.

If a saved version still needs more work to meet your request, Iris can make targeted changes to the presentation. In a report she changes only the key statement and the summary.

Which survey runs Iris analyzes

Iris computes over every answer of a finished survey run, even with very many open answers. While the survey run is still in progress, Iris reports The Run has not finished yet. For a cancelled or failed survey run she computes over the answers given until it stopped and says so beside the numbers. If nobody was surveyed in a run, Iris reports Nobody was surveyed in this Run. No answers will come later either.

Numbers in the text

When Iris refers to a result number from a calculation, the app fills in the number and writes it in your language, for example 43%, +7 pp for a difference in percentage points or 3.5 and 1.49 for a mean, with up to two decimals. If the app can't load a number, — stands in its place.

Iris writes other numbers, such as panel sizes, durations or numbers from your message, as usual.

Small base

If fewer than 30 people stand behind a number, it has a dotted underline. Hover over it or tab to it with the keyboard, and a tooltip shows Small base (n = 7).

Answer from Iris with filled-in numbers, one with a dotted underline and the tooltip Small base (n = 7)
A number with a small base

The limit of 30 people applies to every panel type. The app never hides a number with a small base. Read it as a hint, not as a solid result.

Judging differences

Iris never writes "significant" and gives no p-values or stars, even when you ask for them. All answers are simulated, Premium Twins included. A significance test would suggest a representative sample. Instead, Iris grades a difference in words:

WordingMeaning
clear differenceThe difference is large for the number of people behind it, and both groups have at least 30 people.
slight tendencyThe direction is visible but doesn't carry far, for example because a group is small, the answers are very uniform or the difference is one of many comparisons.
not distinguishable at this group sizeThe difference is small, or a group has fewer than 10 people.

With fewer than 10 people, Iris gives counts instead of percentages, for example "5 of 7". When answers are very uniform, Iris adds that simulated respondents often agree more than real people do.

Quotes

When Iris quotes an answer, the app cuts the wording from the stored answer. Iris never writes quotes herself. Below the quote are attributes of the respondent, such as gender, age group, region and panel type.

If the answer is in a different language from your app, the quote shows the translation with Machine translated. Until the translation arrives, the original appears with Being translated. Show original shows the respondent's wording, Show translation switches back.

Answer from Iris with an indented quote, below it the respondent's attributes, the label Machine translated and Show original
A quote in the chat

If the app can't load a quote, it shows Quote unavailable.

Theme list for open answers

For open questions you can ask Iris for a theme list, for example "Sort the answers to the taste question into themes". Iris reads the answers and writes up to 12 themes, each with a name, a definition and an example. A separate model then assigns every single answer to the themes. One answer can belong to several themes.

In her answer Iris names the themes with their shares, the No theme share and the base. Not evaluable are answers the model couldn't judge. They don't count towards the base.

  • Adjust: Ask Iris to rename a theme, merge two themes, sharpen a definition or assign a single answer differently. Every change saves a new version. Until someone changes the theme list, it gives the same numbers.
  • Across survey runs: If several survey runs ask the same open question, for example in two waves, one theme list can sort them all together.
  • Very many answers: If a question has more than 1,600 answers, Iris forms no theme list and tells you so.

The themes in the report and on a question's card come from the automatic analysis. A theme list in the chat doesn't change them. If you want it in a presentation, ask Iris. If Iris reports The answers could not be assigned right now., try again later.

Computing with files

Iris can also use numbers from your PDFs, Excel and CSV files, such as the market shares from a market study, and put them next to your survey results. Tell Iris which file to use.

  • Iris only computes with PDFs the app has finished preparing. If a PDF is still being prepared, Iris waits briefly.
  • Iris computes Excel and CSV files over every row of the original file, right after the upload.
  • She can't use images, Word and PowerPoint files or JSON for calculations.

More on files in Files.