[[["わかりやすい","easyToUnderstand","thumb-up"],["問題の解決に役立った","solvedMyProblem","thumb-up"],["その他","otherUp","thumb-up"]],[["必要な情報がない","missingTheInformationINeed","thumb-down"],["複雑すぎる / 手順が多すぎる","tooComplicatedTooManySteps","thumb-down"],["最新ではない","outOfDate","thumb-down"],["翻訳に関する問題","translationIssue","thumb-down"],["サンプル / コードに問題がある","samplesCodeIssue","thumb-down"],["その他","otherDown","thumb-down"]],["最終更新日 2025-08-17 UTC。"],[[["\u003cp\u003eMeridian primarily measures the causal effect of treatment variables on a key outcome metric, typically revenue, but can be a different KPI if revenue data is unavailable.\u003c/p\u003e\n"],["\u003cp\u003eReturn on investment (ROI) is conceptually defined as the incremental outcome generated by a media channel divided by its cost, requiring an understanding of causal effects.\u003c/p\u003e\n"],["\u003cp\u003eMeridian uses the concept of potential outcomes from causal inference to define and estimate the incremental outcome generated by different media scenarios.\u003c/p\u003e\n"],["\u003cp\u003eIncremental outcome is defined as the expected difference in outcomes between two counterfactual media scenarios, conditioned on observed control variables.\u003c/p\u003e\n"],["\u003cp\u003eA typical use case involves comparing actual historical media values to a scenario where one specific channel's values are set to zero, allowing for the attribution of incremental outcome and ROI to that channel.\u003c/p\u003e\n"]]],["Meridian uses \"outcome,\" often revenue or a KPI, to measure the causal effect of treatment variables. Return on investment (ROI) is defined as the incremental outcome from a media channel divided by its cost. Incremental outcome, using causal inference, involves comparing potential outcomes under different media scenarios (counterfactuals). It then defines incremental outcome between any two counterfactuals as the expected difference of outcomes conditional on a set of control variables. The expected difference can be estimated through an MMM model with control variables.\n"],null,["# Incremental outcome definition\n\n*Outcome* is the primary metric of interest that Meridian measures the causal\neffect of treatment variables upon. This is typically revenue, but when the KPI\nis not revenue and `revenue_per_kpi` data is not available, then\nMeridian defines the outcome to be the KPI itself.\n\nColloquially, you can define return on investment (ROI) as the incremental\noutcome generated by a media channel divided by its cost. This implies that\nmedia has a causal effect on outcome that you want to estimate. To do this in a\nprincipled way, you need to define *incremental outcome* using the language of\ncausal inference.\n\nFor demonstration purposes, consider the case where no paid or organic media\nchannels have reach and frequency data. Using the notation from [Input\ndata](/meridian/docs/basics/input-data), you have an observed array of\ntransformed media units \\\\(\\\\{x\\^{\\[M\\]}_{g,t,i}\\\\}\\\\), organic media units\n\\\\(\\\\{x\\^{\\[OM\\]}_{g,t,i}\\\\}\\\\), and non-media treatments \\\\(\\\\{x\\^{\\[N\\]}_{g,t,i}\\\\}\\\\)\nthe entirety of which is denoted by \\\\(\\\\{x_{g,t,i}\\\\}\\\\). This set includes\nvalues for all paid and organic media channels, and non-media treatments for\nall \\\\(g=1,\\\\dots G \\\\)\nand \\\\(t=-\\\\infty,\\\\dots,T \\\\), although in practice you only\nneed to worry about \\\\(t=1-L,2-L,\\\\dots T\\\\) where \\\\(L\\\\) is the assumed maximum\nlag of media effects. (For the purposes of this discussion, refer to units on\nthe transformed scale \\\\(x_{g,t,i}\\\\) instead of the raw scale \\\\(\\\\overset{\\\\cdot\n\\\\cdot}{x}_{g,t,m}\\\\). There is a one-to-one correspondence between raw and\ntransformed units, so it makes no practical difference.)\n\nEven if an advertiser's actual execution was \\\\(\\\\{x_{g,t,i}\\\\}\\\\), you can\nimagine what the outcome might have been if the advertiser had instead\nexecuted a different media array, such as \\\\(\\\\{x\\^{(\\\\ast)}_{g,t,i}\\\\}\\\\). You can\ndenote this outcome as the set of random variables \\\\(\\\\{ \\\\overset \\\\sim\nY_{g,t}\\^{ (\\\\{ x_{g,t,i}\\^{(\\\\ast)} \\\\}) }\\\\}\\\\). In the causal inference literature,\nthe set \\\\(\\\\{ \\\\overset \\\\sim Y_{g,t}\\^{ (\\\\{ x_{g,t,i}\\^{(\\\\ast)} \\\\}) }\\\\}\\\\) are\ncalled *potential outcomes* , and the set of values \\\\(\\\\{ x\\^{(\\\\ast)}_{g,t,i}\n\\\\}\\\\) is called a *counterfactual scenario*.\n\nIn the causal inference literature, it is common to see notation like\n\\\\(Y\\^{(1)}\\\\) and \\\\(Y\\^{(0)}\\\\) representing potential outcomes under\n*treatment* and *control* counterfactual scenarios. MMM is similar but slightly\nmore complex because the potential outcomes are a two-dimensional array of\nvalues, and the treatment is a three-dimensional array of values. Note that not\nevery potential outcome in the array \\\\(\\\\{ \\\\overset \\\\sim Y_{g,t}\\^{ (\\\\{\nx_{g,t,i}\\^{(\\\\ast)} \\\\}) }\\\\}\\\\) actually depends on all values in the array \\\\(\\\\{\nx\\^{(\\\\ast)}_{g,t,i} \\\\}\\\\). For example, media in a given time period cannot\naffect past sales. However, this notation is preferred because it is simpler\nthan trying to denote exactly which media values each potential outcome depends\non for each time period.\n\nAlthough for any two counterfactual media scenarios, such as \\\\(\\\\{\nx\\^{(1)}_{g,t,i} \\\\}\\\\) and \\\\(\\\\{ x\\^{(0)}_{g,t,i} \\\\}\\\\), you could define the\nactual incremental outcome as: \n$$ \\\\sum\\\\limits _{g,t} \\\\left( \\\\overset \\\\sim Y_{g,t}\\^{ \\\\left( \\\\left\\\\{ x_{g,t,i}\\^{(1)} \\\\right\\\\} \\\\right) } - \\\\overset \\\\sim Y_{g,t}\\^{ \\\\left( \\\\left\\\\{ x_{g,t,i}\\^{(0)} \\\\right\\\\} \\\\right) } \\\\right) $$\n\nHowever, this quantity is not estimable because data cannot provide any\ninformation about the joint distribution of \\\\(\\\\overset \\\\sim Y_{g,t}\\^{ (\\\\{\nx_{g,t,i}\\^{(1)} \\\\}) }\\\\) and \\\\(\\\\overset \\\\sim Y_{g,t}\\^{ (\\\\{ x_{g,t,i}\\^{(0)} \\\\})\n}\\\\). It is only possible to observe one potential outcome, namely \\\\(\\\\overset\n\\\\sim Y_{g,t}\\^{ \\\\left( \\\\left\\\\{ x_{g,t,i} \\\\right\\\\}\\\\right) }\\\\). (Note that\nintuitively, as \\\\(\\\\{ x\\^{(1)}_{g,t,i} \\\\}\\\\) becomes arbitrarily close to \\\\(\\\\{\nx\\^{(0)}_{g,t,i} \\\\}\\\\), the potential outcomes \\\\(\\\\overset \\\\sim Y_{g,t}\\^{ (\\\\{\nx_{g,t,i}\\^{(1)} \\\\}) }\\\\) and \\\\(\\\\overset \\\\sim Y_{g,t}\\^{ (\\\\{ x_{g,t,i}\\^{(0)} \\\\})\n}\\\\) should approach the same value, but this intuition is not enough to specify\nthe joint distribution more generally.)\n\nInstead, for any two counterfactual media scenarios, \\\\(\\\\{\nx\\^{(1)}_{g,t,i} \\\\}\\\\) and \\\\(\\\\{ x\\^{(0)}_{g,t,i} \\\\}\\\\), define incremental\noutcome as: \n$$ \\\\text{IncrementalOutcome} \\\\left( \\\\left\\\\{ x\\^{(1)}_{g,t,i} \\\\right\\\\}, \\\\left\\\\{ x\\^{(0)}_{g,t,i} \\\\right\\\\} \\\\right) = E \\\\left( \\\\sum\\\\limits_{g,t} \\\\left( \\\\overset \\\\sim Y_{g,t}\\^{ \\\\left(\\\\left\\\\{ x_{g,t,i}\\^{(1)} \\\\right\\\\}\\\\right) } - \\\\overset \\\\sim Y_{g,t}\\^{ \\\\left(\\\\left\\\\{ x_{g,t,i}\\^{(0)} \\\\right\\\\}\\\\right) } \\\\right) \\\\Bigg\\| \\\\left\\\\{ z_{g,t,i} \\\\right\\\\} \\\\right) $$\n\nwhere \\\\(\\\\{z_{g,t,i}\\\\}\\\\) denotes the observed values for a set of control\nvariables. This shorthand notation is used to indicate that the expectation is\nconditional upon the control random variables taking on these values. Using an\nMMM regression model and a carefully selected set of control variables, this\nconditional expectation is estimable. For more information, see [ROI, mROI, and\nresponse curves](/meridian/docs/basics/roi-mroi-response-curves).\n\nTypically, the sum is taken over \\\\(g=1,\\\\dots G\\\\) and \\\\(t=1,\\\\dots T\\\\),\nhowever, you can also define incremental outcome for any subset of these values.\n| **Note:** A common pair of counterfactual media scenarios used in Meridian is to let $x\\^{(1)}$ be actual historical values and $x\\^{(0)}$ be the same as $x\\^{(1)}$ except with all values for one specific channel set to zero. The resulting incremental outcome is defined as the incremental outcome attributed to that channel, which is also the ROI numerator for that channel."]]