Using Holdout Control Group Feature for SAS 360 Engage Direct Marketing Tasks

/* Define control group percentage */
%let control_pct = 10;  /* 10% holdout */

/* Randomly assign control status */
data campaign_audience;
  set target.customers;
  
  /* Random assignment */
  if ranuni(12345) <= (&control_pct / 100) then 
    control_group = 1;
  else 
    control_group = 0;
  
  /* Ensure representation across segments */
  call streaminit(12345);
run;

/* Stratified control group */
proc surveyselect data=target.customers
  out=stratified_control
  method=srs
  rate=0.10  /* 10% control */
  seed=12345;
  strata segment;  /* Stratify by segment */
run;

/* Matched control group design */
proc psmatch data=customers;
  class gender region (ref='WEST');
  psmodel treatment(treated='1') = 
    age gender region income;
  match method=greedy(k=1) 
    distance=lps 
    caliper=0.25;
  assess lps / plots=all;
  output out=matched_groups atc=_atc_;
run;

/* Create control based on propensity matching */
data control_matched;
  set matched_groups;
  where _matchwgt_ > 0 and treatment = 0;
  control_group = 1;
run;

/* Prepare campaign audience (excluding controls) */
data campaign_send_list;
  set campaign_audience;
  where control_group = 0;
  
  /* Additional campaign fields */
  campaign_id = "CAMP_2024_Q1";
  send_date = today();
  channel = "EMAIL";
run;

/* Log control group for analysis */
data campaign_control_log;
  set campaign_audience;
  where control_group = 1;
  
  campaign_id = "CAMP_2024_Q1";
  control_date = today();
  
  /* Track for future analysis */
run;

/* Append to historical controls */
proc append base=history.control_groups
  data=campaign_control_log;
run;

/* Analyze campaign vs control */
proc sql;
  create table campaign_results as
  select 
    case when control_group = 1 then 'Control' else 'Treatment' end as group,
    count(*) as customers,
    sum(purchased) as purchasers,
    sum(revenue) as total_revenue,
    calculated total_revenue / calculated customers as avg_revenue
  from campaign_audience a
  left join transactions t
  on a.customer_id = t.customer_id
    and t.trans_date between campaign_date and campaign_date + 30
  group by calculated group;
quit;

/* Calculate lift */
data campaign_lift;
  set campaign_results;
  by group;
  
  retain control_rate control_revenue;
  
  if group = 'Control' then do;
    control_rate = purchasers / customers;
    control_revenue = avg_revenue;
  end;
  else do;
    lift_rate = (purchasers / customers - control_rate) / control_rate * 100;
    lift_revenue = (avg_revenue - control_revenue) / control_revenue * 100;
    incremental_revenue = (avg_revenue - control_revenue) * customers;
  end;
run;

/* Report results */
proc print data=campaign_lift noobs;
  var group customers purchasers total_revenue avg_revenue 
      lift_rate lift_revenue incremental_revenue;
  format total_revenue incremental_revenue dollar15.2
         avg_revenue dollar10.2
         lift_rate lift_revenue percent8.1;
run;

/* Test for significant differences */
proc ttest data=campaign_audience;
  class control_group;
  var revenue;
  title "Revenue Difference: Treatment vs Control";
run;

/* Chi-square test for conversion */
proc freq data=campaign_audience;
  tables control_group * purchased / chisq;
  title "Conversion Rate Comparison";
run;

/* Documented control group creation */
%macro create_control_group(
  input_data=,
  output_data=,
  control_pct=10,
  strata_vars=,
  random_seed=12345
);
  
  %put NOTE: Creating control group;
  %put NOTE: Control percentage: &control_pct;
  %put NOTE: Stratification variables: &strata_vars;
  %put NOTE: Random seed: &random_seed;
  
  %if &strata_vars ne %then %do;
    /* Stratified sampling */
    proc surveyselect data=&input_data
      out=&output_data
      method=srs
      rate=%sysevalf(&control_pct/100)
      seed=&random_seed;
      strata &strata_vars;
    run;
  %end;
  %else %do;
    /* Simple random sampling */
    data &output_data;
      set &input_data;
      if ranuni(&random_seed) <= (&control_pct / 100) then
        control_group = 1;
      else
        control_group = 0;
    run;
  %end;
  
  /* Verify control group size */
  proc sql noprint;
    select count(*) * 100.0 / 
      (select count(*) from &input_data)
    into :actual_pct
    from &output_data
    where control_group = 1;
  quit;
  
  %put NOTE: Actual control percentage: &actual_pct;
  
%mend;

/* Usage */
%create_control_group(
  input_data=target.customers,
  output_data=campaign.audience,
  control_pct=10,
  strata_vars=segment region,
  random_seed=12345
);

/* Automated control group analysis report */
%macro control_report(campaign_id=);
  ods pdf file="control_analysis_&campaign_id..pdf";
  
  title "Control Group Analysis: &campaign_id";
  
  /* Summary statistics */
  proc means data=results;
    class control_group;
    var revenue conversion_rate;
  run;
  
  /* Lift calculations */
  proc print data=lift_summary;
  run;
  
  /* Visualizations */
  proc sgplot data=results;
    vbar control_group / response=revenue stat=mean;
    title2 "Average Revenue by Group";
  run;
  
  ods pdf close;
%mend;