AB Test API Version 3.1 - Documentation

AB Test API Version 3.1 - Docs

Bayesian Chance to Beat into Google Sheets

In the latest version 3.1 (December 2021): In the pre-latest version 3.0 (April 2021): Please refer to the updated example sheet for all options.

Introduction

The Bayesian API at abtestapi.io provides an endpoint that returns the Bayesian "Chance to Beat" scores for A/B/n tests. The primary use case is integrating Bayesian results into Google Sheets (using the "IMPORTDATA" function). By default the output with be returned in .csv (download) format, since this is required for the Google Sheet ImportData function.

JSON formatted output for other uses is also possible by using the 'output=json' parameter.

Access / usage limits

Access is granted by using a company-specific token that is provided to you by Online Dialogue. Please note requests without a valid token will no longer work.



Input

Calling the url 'https://abtestapi.io/bayes', followed by required query parameter token with your company-specific token, and parameters a & b (required) for users and conversion in the control group (separated by '|' pipe symbol). Optionally additional variations can be added using c, d, ... j parameters.

Output

The url request will return a .csv formatted result, which can be used in the import function in Google Sheets.
The output will show a chance between 0 and 1 for each of the variants, with 4 digits accuracy. The number of samples and output format can be configured (see API Reference).

Examples

Title Description
URL (json) https://abtestapi.io/bayes?token=[YourTokenHere]&a=3010|153&b=3020|153&output=json
URL (csv) https://abtestapi.io/bayes?token=[YourTokenHere]&a=3010|153&b=3020|153
Google Sheets Formula =IMPORTDATA("https://abtestapi.io/bayes?token=[YourTokenHere]&a="&B2&"|"&C2&"&b="&B3&"|"&C3)
Output options Google Sheet showing various output options.

API Reference Bayesian

Base URL: https://abtestapi.io/bayes

Parameter Type Options
token required Unique company-specific token, provided to you by [email protected].
Please note requests without a valid token will no longer work.
a required Users & Conversions in A: Control, separated by a pipe '|' symbol.
For example for 3010 users with 153 conversions: a=3010|153
b required Users & Conversions in B: Variant, separated by a pipe '|' symbol.
For example: b=3020|164
c, d, e .. j optional Additional variants for AB Test with more than 2 variants. Same format as a, b.
output optional For information on all output options, please refer to this example sheet.
decSep optional Decimal separator, 'comma' or 'dot' (default='dot'). When 'comma' is selected, the csv is suitable for import in a 'European' locale Google Sheets that uses comma as decimal separator. The csv delimiter then changes to semicolon; beware to use the undocumented second argument of ";" in the IMPORTDATA function.
samples optional Number of samples to use for the Monte Carlo Bayesian inference. Default = 50000. The higher the number, the longer processing time it will take. Maximum 5 million samples is allowed.
header optional true or false.
Show a header row with column names.
intervalSignificance optional Significance of the confidence interval of the difference. Default = 0.95
srmThreshold optional Threshold (p-value) for the Sample Ratio Mismatch check. Default = 0.00005
Business Case feature (only applicable when Business Case results are required)
aov required Average Order Value. Required only when outputing Business Case results.
duration required Duration of the Test in days
percInTest optional Percentage traffic in test. Default 100%
effectDuration optional How long uplift effect of variation will sustain, in months. Default 6 months.

API Reference MDE API

Base URL: https://abtestapi.io/mde

Parameter Type Options
token required Unique company-specific token, provided to you by [email protected].
Please note requests without a valid token will no longer work.
traffic required Weekly total number of users that are exposed to the control or variation.
conversions required Weekly total number of users with a conversion, that are exposed to the control or variation.
power optional Statistical Power (decimal '.' dot-separated). Default: 0.8
sign optional Statistical Significance (decimal '.' dot-separated). Default: 0.95
sides optional Two ('2') or one ('1') sided test. Default: 1
weeks optional Number of weeks to return results for. Default: 1