Skip to contents

Check the data

Find collinear, flat or dark channels and inconsistent costs before fitting anything.

diagnose_media()
Diagnose media data before modelling
plot(<mm_diagnosis>)
Plot media diagnostics

Carryover (adstock)

Spread each period’s media forward in time, and translate decay into a half-life.

adstock_geometric()
Geometric adstock
adstock_weibull()
Weibull adstock
adstock_weights_delayed() adstock_delayed()
Delayed adstock
adstock_filter()
Adstock with an arbitrary kernel
adstock_weights()
Geometric adstock weights
adstock_weights_weibull()
Weibull adstock weights
adstock_state()
Terminal adstock state
adstock_steady_state()
Steady-state starting value for an adstock filter
decay_from_half_life() half_life() effective_window()
Carryover decay vocabulary

Saturation

Diminishing returns as media pressure grows.

Composing transforms

Apply carryover then saturation to one channel or to every channel of a plan, and convert Robyn’s hyperparameters to these arguments.

media_transform()
Adstock then saturate, in that order
transform_plan()
Transform every channel of a media plan
from_robyn()
Convert Robyn hyperparameters to mediamix arguments

Choosing carryover parameters

Score candidate decays and lags by how well a model predicts held-out KPI, with seasonality terms as controls.

tune_carryover()
Select carryover parameters by cross-validation against a KPI
tune_carryover_joint()
Tune every channel's carryover jointly, inside one model
plot(<mm_carryover>) plot(<mm_carryover_joint>)
Plot a carryover cross-validation profile
fit_ols() predict(<mm_ols>) coef(<mm_ols>) fitted(<mm_ols>) print(<mm_ols>)
Default linear fitter for carryover tuning
rmse() mae()
Root mean squared error and mean absolute error
fourier_terms() holiday_flags()
Seasonality controls for a model frame

Tuning with tidymodels

Recipe steps and tuning parameters for tuning carryover, saturation and a model penalty together.

step_adstock()
Adstock transformation as a recipe step
step_adstock_channels()
One adstock step per channel
step_saturation()
Saturation transformation as a recipe step
carryover_decay() carryover_max_lag() carryover_shape() carryover_scale() carryover_peak() saturation_half_max() saturation_shape()
Tuning parameters for media transforms

Contributions and return on investment

Turn a fitted model into channel contributions, average and marginal return, and response curves.

contributions()
Decompose predicted KPI into channel contributions
plot(<mm_contributions>)
Plot a contribution decomposition
roi() mroi()
Return on investment and marginal return
marginal_roi()
Average and marginal return by counterfactual simulation
response_curve() spend_for()
Response curve for a channel
plot(<mm_response_curve>)
Plot a response curve

Uncertainty

Put intervals on any statistic by resampling blocks of consecutive periods.

block_bootstrap()
Moving block bootstrap for time-series statistics
plot(<mm_bootstrap>)
Plot bootstrap intervals

Budget planning

Simulate a plan through the fitted transforms, move budget between channels, and split a total budget across channels.

simulate_plan()
Predicted media response of a plan
reallocate()
Move spend between two channels of a plan
allocate_budget()
Allocate a budget across channels
plot(<mm_allocation>)
Plot a budget allocation

Journey construction

Turn a raw event log into customer journeys, and combine or subset them.

build_paths()
Build customer journeys from a raw event log
bind_paths() rbind(<mm_paths>)
Combine journey tables
`[`(<mm_paths>)
Subset a journey table

Journey diagnostics

Describe a journey table, including how long conversions take, before choosing a lookback or a credit rule.

path_summary()
Journey table summary
path_lengths()
Journey length distribution
channel_positions()
Where each channel sits in the journey
assisted_conversions()
Assisted conversion matrix
top_paths()
Most common journeys
conversion_lag()
Conversion lag distribution
path_diagnostics()
Run every journey diagnostic at once

Attribution

Credit conversions to channels by rule or by Markov removal effects, compare the rules, and export journeys to ChannelAttribution.

credit_linear() credit_first() credit_last() credit_position() credit_time_decay() credit_custom()
Assign conversion credit across a journey
markov_removal()
Markov-chain removal effects
attribute()
Attribute conversions to channels under several rules at once
attribution_spread()
Range of each channel's share across attribution rules
plot(<mm_attribution>)
Plot attribution shares across rules
plot(<mm_attribution_intervals>) plot(<mm_attribution_segments>)
Plot attribution shares with intervals or by segment
as_channel_paths()
Export journeys in ChannelAttribution's format

Example data

A simulated weekly media panel and event log, generated from known parameters so results can be checked.

mm_weekly
Synthetic weekly marketing mix panel
mm_events
Synthetic touchpoint event log

Package

The package overview, and the colour palette its plots use.

mediamix mediamix-package
mediamix: media transforms and attribution path construction
mm_palette()
The package's chart palette