Methods, systems, apparatus and articles of manufacture to determine causal effects

Abstract

Methods, systems, apparatus, and articles of manufacture to determine causal effects are disclosed herein. An example apparatus includes a weighting engine to calculate a first set of weights corresponding to a first treatment dataset, a second set of weights corresponding to a second treatment dataset, and a third set of weights corresponding to a control dataset, the weighting engine to increase an operational efficiency of the apparatus by calculating the first set of weights, second set of weights, and third set of weights independently, a weighting response engine to calculate a first weighted response for the first treatment dataset, a second weighted response for the second treatment dataset, and determine a causal effect between the first treatment dataset and the second treatment dataset based on a difference between the first weighted response and the second weighted response, and a report generator to transmit a report to an audience measurement entity.

What is claimed is: 
     
         1 . An apparatus comprising:
 memory;   instructions; and   processor circuitry to execute the instructions to at least:
 determine a covariate associated with first covariates and second covariates to balance between a first dataset and a second dataset, the first dataset including the first covariates, the second dataset including the second covariates, the first dataset indicative of first individuals who have been exposed to an advertisement, the second dataset indicative of second individuals who have not been exposed to the advertisement; 
 compute, via maximum entropy, first weights for the first covariates and second weights for the second covariates, the first weights and the second weights, when applied to the first covariates and the second covariates, respectively, to cause the covariate associated with the first covariates and the second covariates to be equal between the first covariates and the second covariates; 
 compute a first weighted response for the first dataset based on the first weights and a second weighted response for the second dataset based on the second weights; and 
 determine an effect of the advertisement based on a difference between the first weighted response and the second weighted response. 
   
     
     
         2 . The apparatus of  claim 1 , wherein the processor circuitry is to access the first dataset and the second dataset from a data store. 
     
     
         3 . The apparatus of  claim 1 , wherein to determine the covariate to balance, the processor circuitry is to:
 identify the first dataset as a treatment group dataset and the second dataset as a control group dataset;   identify the first covariates as present in the treatment group dataset and the second covariates present in the control group dataset;   identify at least one covariate that is included in both the treatment group dataset and the control group dataset; and   identify the at least one covariate as the covariate to balance.   
     
     
         4 . The apparatus of  claim 1 , wherein the processor circuitry is to generate a report including the effect, the report configured to be displayed via a webpage. 
     
     
         5 . The apparatus of  claim 1 , wherein the first weighted response is on a common scale with the second weighted response. 
     
     
         6 . The apparatus of  claim 1 , wherein the effect of the advertisement is indicative of an average monetary change in purchases by the first individuals. 
     
     
         7 . The apparatus of  claim 1 , wherein:
 the first covariates include at least one of respective first ages, first genders, first incomes, or first races of the first individuals; and   the second covariates include at least one of respective second ages, second genders, second incomes, or second races of the second individuals.   
     
     
         8 . A non-transitory computer readable medium comprising instructions that, when executed, cause processor circuitry to at least:
 determine a covariate to balance between a first set of covariates of a first dataset and a second set of covariates of a second dataset, the first dataset representative of a first set of individuals who have been exposed to a treatment, the second dataset representative of a second set of individuals who have not been exposed to the treatment;   compute, via maximum entropy, a first set of weights for the first set of covariates and a second set of weights for the second set of covariates, the first set of weights and the second set of weights, when applied to the first set of covariates and the second set of covariates, respectively, to cause the covariate to be equal between the first set of covariates and the second set of covariates;   compute a first weighted response for the first dataset based on the first set of weights and a second weighted response for the second dataset based on the second set of weights;   determine an effect of the treatment based on a difference between the first weighted response and the second weighted response; and   generate a report including the effect, the report configured to be displayed via a webpage.   
     
     
         9 . The non-transitory computer readable medium of  claim 8 , wherein the instructions cause the processor circuitry to access, from a data store, the first dataset and the second dataset. 
     
     
         10 . The non-transitory computer readable medium of  claim 8 , wherein to determine the covariate to balance, the instructions cause the processor circuitry to:
 identify the first dataset as a treatment group dataset and the second dataset as a control group dataset;   identify the first set of covariates as present in the treatment group dataset and the second set of covariates present in the control group dataset;   identify at least one covariate that is included in both the treatment group dataset and the control group dataset; and   identify the at least one covariate as the covariate to balance.   
     
     
         11 . The non-transitory computer readable medium of  claim 8 , wherein the treatment includes at least one of an advertisement, a drug, a tweet, or a product purchase instance. 
     
     
         12 . The non-transitory computer readable medium of  claim 8 , wherein the first weighted response and the second weighted response share a compatible unit of measure. 
     
     
         13 . The non-transitory computer readable medium of  claim 8 , wherein the effect of the treatment is indicative of a causal effect of the treatment. 
     
     
         14 . The non-transitory computer readable medium of  claim 8 , wherein:
 the first set of covariates includes at least one of a first set of ages, a first set of genders, a first set of incomes, or a first set of races of the first set of individuals; and   the second set of covariates includes at least one of a second set of ages, a second set of genders, a second set of incomes, or a second set of races of the second set of individuals.   
     
     
         15 . A method comprising:
 determining, by executing an instruction with processor circuitry, a covariate included in first covariates and second covariates to balance between a first dataset and a second dataset, the first covariates including first demographic characteristics of first individuals represented by the first dataset, the second covariates including second demographic characteristics of second individuals represented by the second dataset, the first individuals having been exposed to an advertisement, the second individuals not having been exposed to the advertisement;   computing, by executing an instruction with the processor circuitry and via maximum entropy, first weights corresponding to the first covariates and second weights corresponding to the second covariates, the first weights and the second weights, when applied to the first covariates and the second covariates, respectively, to equalize the covariate included in the first covariates and the second covariates between the first covariates and the second covariates;   computing, by executing an instruction with the processor circuitry, a first weighted response for the first dataset based on the first weights and a second weighted response for the second dataset based on the second weights; and   determining, by executing an instruction with the processor circuitry, an effect of the advertisement based on a difference between the first weighted response and the second weighted response.   
     
     
         16 . The method of  claim 15 , further including accessing a data store to retrieve the first dataset and the second dataset. 
     
     
         17 . The method of  claim 15 , wherein determining the covariate to balance includes:
 identifying the first dataset as a treatment group dataset and the second dataset as a control group dataset;   identifying the first covariates as present in the treatment group dataset and the second covariates present in the control group dataset;   identifying at least one covariate that is included in both the treatment group dataset and the control group dataset; and   identifying the at least one covariate as the covariate to balance.   
     
     
         18 . The method of  claim 15 , further including generating a report to be displayed via a webpage, the report including the effect. 
     
     
         19 . The method of  claim 15 , wherein the first weighted response shares a common unit of measure with the second weighted response. 
     
     
         20 . The method of  claim 15 , wherein the effect of the advertisement is indicative of an average monetary change in purchases by the first individuals. 
     
     
         21 . The method of  claim 15 , wherein:
 the first demographic characteristics include at least one of respective first ages, first genders, first incomes, or first races of the first individuals; and   the second demographic characteristics include at least one of respective second ages, second genders, second incomes, or second races of the second individuals.