加入新推出的
Discord 社区,展开实时讨论,获得同行支持,并直接与 Meridian 团队互动!
自然媒体和非媒体处理变量
使用集合让一切井井有条
根据您的偏好保存内容并对其进行分类。
除了付费媒体外,可能还会有其他营销行动影响相关 KPI。
自然媒体变量是指不会产生直接费用的媒体活动。这些活动包括但不限于简报、博文、社交媒体活动或电子邮件宣传活动的展示。自然媒体变量可以与覆盖面和频次数据一起纳入到模型中,并且与付费媒体变量一样,具有 Adstock 和 Hill 效应。自然媒体和付费媒体之间的唯一区别在于,自然媒体没有关联的费用。因此,投资回报率先验无法与自然媒体配合使用,并且模型不会针对自然媒体变量提供与投资回报率相关的结果,例如响应曲线和预算优化。
模型会针对自然媒体变量提供因果效应和贡献百分比,并且这两者的计算方式与付费媒体相同,如增量效果定义中所述。与付费媒体一样,自然媒体渠道的增量效果定义为:与不运行所观测渠道的反事实情景相比,在所观测渠道的媒体执行下效果的预期差异。
非媒体处理变量是指与媒体没有直接关系的营销活动,例如开展促销活动、调整产品价格以及改变产品包装或设计。这类变量没有关联的直接营销费用,但与自然媒体变量不同,它们与媒体无关,也没有 Adstock 和 Hill 效应。它们与控制变量不同,因为它们被视为可干预的变量,因此在因果模型中属于处理变量。鉴于此,模型会针对非媒体变量提供增量效果和贡献百分比。
与付费媒体和自然媒体类似,非媒体变量的增量效果定义为两个反事实情景之间的预期效果差异。第一个情景是将非媒体变量设置为每个地理位置和时间段的历史观测值。第二个情景是将非媒体变量设置为非媒体变量的最小值(默认)、最大值或用户提供的值(请参阅 ModelSpec
中的 non_media_baseline_values
实参)。增量效果就是第一个情景下的预期效果减去反事实情景下的预期效果。
第二个情景(即反事实情景)之所以没有将非媒体变量设为零(就像在付费媒体和自然媒体中那样),是因为对于非媒体变量,零通常不是一个适当的反事实值。例如,如果非媒体变量是价格,那么分别将价格设为观测值和产品最低销售价格,通过比较来了解因果效应可能是合理的,但将价格设为零则没有意义。
非媒体变量与控制变量的主要区别在于,非媒体变量被视为可干预的变量,并因此在假设的因果模型中属于处理变量。控制变量(也称为混杂变量)无法干预,并且模型假设它们会同时影响处理变量和结果。如需了解详情,请参阅因果图。
作为广告客户,如果您可以干预并更改某个变量的值(例如更改价格或开展促销活动),那么该变量更有可能是非媒体变量,而不是控制变量。如果该变量不在广告客户的控制范围内,例如广义经济指标或地理位置级/国家级受众特征,则很可能是控制变量。
自然媒体变量的行为类似于没有关联费用的付费媒体变量。它们一般基于展示或覆盖面和频次,通常是没有直接费用的广告活动,例如社交媒体帖子和电子邮件宣传活动。非媒体变量也没有直接费用,但与媒体无关。通常,非媒体变量与基础产品的变化相关,例如价格、促销活动或产品包装的变化。另一种判断哪种变量适当的方法是,自然媒体变量会应用 Adstock 和 Hill 效应,但非媒体变量不会。
下表有助于您确定哪种输入变量合适:
输入变量 |
费用 |
Adstock/Hill |
可干预 |
效应(贡献百分比) |
media |
x |
x |
x |
x |
non_media |
- |
- |
x |
x |
organic_media |
- |
x |
x |
x |
controls |
- |
- |
- |
- |
如需详细了解 Meridian 中输入变量之间的假设因果关系,请参阅因果图。
如未另行说明,那么本页面中的内容已根据知识共享署名 4.0 许可获得了许可,并且代码示例已根据 Apache 2.0 许可获得了许可。有关详情,请参阅 Google 开发者网站政策。Java 是 Oracle 和/或其关联公司的注册商标。
最后更新时间 (UTC):2025-08-08。
[null,null,["最后更新时间 (UTC):2025-08-08。"],[[["\u003cp\u003eMeridian MMM supports analyzing the impact of organic media and non-media marketing activities alongside paid media.\u003c/p\u003e\n"],["\u003cp\u003eOrganic media activities, like social media and email campaigns, have no direct cost but can be modeled with reach, frequency, Adstock, and Hill effects.\u003c/p\u003e\n"],["\u003cp\u003eNon-media activities, such as price changes or promotions, are not directly media-related and do not have Adstock or Hill effects.\u003c/p\u003e\n"],["\u003cp\u003eBoth organic media and non-media activities are considered intervenable and contribute to the overall marketing impact.\u003c/p\u003e\n"],["\u003cp\u003eThe appropriate categorization of a variable depends on factors like cost, media relation, and whether it can be controlled by the advertiser.\u003c/p\u003e\n"]]],["The document outlines three types of marketing variables: organic media, non-media, and paid media, alongside control variables. Organic media, like social posts, have no direct cost but include reach, frequency, Adstock, and Hill effects; ROI metrics are not available. Non-media variables, such as promotions or price changes, lack direct costs, and Adstock/Hill effects, but have calculable incremental impacts. They are intervenable. Control variables are non-intervenable factors that affect both the treatments and the outcome.\n"],null,["# Organic media and non-media treatment variables\n\nIn addition to paid media there may be other marketing actions taken place that\naffect the KPI of interest.\n\nOrganic media variables\n-----------------------\n\nOrganic media variables are media activities that have no direct cost. These can\ninclude, but are not limited to, impressions from newsletters, a blog post,\nsocial media activity, or email campaigns. Organic media variables have the\noption of being modeled with reach and frequency, and have Adstock and Hill\neffects, just like paid media variables. The only difference between organic and\npaid media is that organic media does not have an associated cost. So, ROI\npriors cannot be used with organic media and results pertaining to ROI, such as\nresponse curves and budget optimization, are not available for organic media\nvariables.\n\nCausal effects and % (percent) contributions are given for organic media\nvariables and are calculated in the same manner as paid media, as described in\n[Incremental outcome\ndefinition](/meridian/docs/basics/incremental-outcome-definition). As with paid\nmedia, the incremental outcome of an organic media channel is defined as the\nexpected difference in the outcome under the observed channel's media execution\ncompared with the counterfactual scenario of not running that channel.\n\nNon-media treatment variables\n-----------------------------\n\nNon-media treatment variables are marketing activities that are not directly\nrelated to media, such as running a promotion, the price of a product, and a\nchange in a product's packaging or design. They have no direct marketing cost\nassociated with them but, unlike organic media variables, they are not media\nrelated and there are no Adstock and Hill effects. They differ from control\nvariables because they are considered to be intervenable and therefore are\ntreatment variables under the causal model. As such, incremental outcome and %\n(percent) contributions are provided for non-media variables.\n\nSimilar to paid and organic media, the incremental outcome of a non-media\nvariable is defined as the difference in expected outcome between two\ncounterfactual scenarios. The first scenario is setting the non-media variable\nto the observed historical value for each geo and time period. The second\nscenario is setting the non-media variable to either the minimum of the\nnon-media variable (default), the maximum or a user supplied value (see\n`non_media_baseline_values` argument in `ModelSpec`). The incremental outcome\nis the expected outcome under the first scenario subtracted by the expected\noutcome under the counterfactual scenario.\n\nThe reason the second counterfactual is not setting the non-media variable to\nzero, as is the case with paid and organic media, is that zero is often not an\nappropriate counterfactual for non-media variables. For example, if the\nnon-media variable is price it may make sense to think of the causal effect of\nsetting the price to its observed value compared with the minimum price the\nproduct sold for but it does not make sense to set the price to zero.\n\nDeciding if a variable is a non-media treatment variable or a control\n---------------------------------------------------------------------\n\nThe main difference between a non-media variable and a control is that non-media\nvariables are considered to be intervenable, and therefore are treatments under\nthe assumed causal model. Controls, otherwise known as confounding variables,\nare not intervenable and are assumed to affect both the treatment variables and\nthe outcome. For more information, see [Causal\nGraph](/meridian/docs/basics/causal-graph).\n\nAs an advertiser, if you can intervene and change a variable's value (such as\nchanging the price or running a promotion), the variable is more likely a\nnon-media variable than a control. If the variable is outside the control of the\nadvertiser, such as, broad economic indicators or geo or national level\ndemographics then it is most likely a control.\n\nDeciding if a variable is an organic media variable or a non-media treatment variable\n-------------------------------------------------------------------------------------\n\nOrganic media variables behave like paid media variables with no associated\ncost. They usually are impression or reach and frequency based, and are\ntypically advertising activities with no direct cost, such as social media posts\nand email campaigns. Non-media variables also have no direct cost but are not\nmedia related. Typically, non-media variables are related to changes in the\nunderlying product, such as the price, a promotion or a change in the product's\npackaging. Another way to determine what is appropriate for your variable is\nthat organic media have Adstock and Hill effects applied, but non-media\nvariables don't.\n\nDifferences between the types of input variables\n------------------------------------------------\n\nThe following table can help determine what input variable is appropriate:\n\n| Input Variable | Cost | Adstock/Hill | Intervenable | Effect (% Contribution) |\n|-----------------|------|--------------|--------------|-------------------------|\n| `media` | x | x | x | x |\n| `non_media` | - | - | x | x |\n| `organic_media` | - | x | x | x |\n| `controls` | - | - | - | - |\n\nFor more information about assumed causal relationships between input variables\nin Meridian, see [Causal Graph](/meridian/docs/basics/causal-graph)."]]