Measurement triangulation

측정 삼각검증

Compare attribution, MMM, and lift evidence to investigate disagreement.

···
html
<div class="tri"><div class="head">ONE CHANNEL · THREE LENSES <b id="spread">SPREAD: WIDE</b></div><div class="scale"><div class="band" id="band"></div><div class="mark" id="m1"><span>ATTRIBUTION</span><i></i></div><div class="mark" id="m2"><span>MMM</span><i></i></div><div class="mark" id="m3"><span>LIFT TEST</span><i></i></div></div><div class="base"><span>LOW EFFECT</span><span>HIGH EFFECT</span></div></div>
css
.tri{width:min(86%,600px);font:850 clamp(10px,max(2vmin,1vw),14px)/1.2 sans-serif}.head,.base{display:flex;justify-content:space-between;gap:6px}.head{border-bottom:2px solid var(--fg);padding-bottom:8px}.head b{color:var(--accent)}.scale{position:relative;height:clamp(90px,25vmin,180px);border-bottom:2px solid var(--fg);margin-top:8px}.band{position:absolute;left:20%;width:60%;height:70%;bottom:0;background:color-mix(in srgb,var(--accent) 17%,transparent);transition:left .65s,width .65s}.mark{position:absolute;left:20%;width:1px;height:clamp(52px,13vmin,95px);bottom:0;background:var(--fg);transition:left .65s}.mark span{position:absolute;bottom:100%;left:50%;transform:translateX(-50%);white-space:nowrap;font-size:clamp(10px,max(1.5vmin,1vw),12px)}.mark i{position:absolute;bottom:0;left:-5px;width:11px;height:11px;border-radius:50%;background:var(--accent)}.mark:nth-child(3) i{background:var(--accent-2)}.mark:nth-child(4) i{background:var(--accent-3)}.base{padding-top:7px;color:var(--muted);font-size:clamp(10px,max(1.5vmin,1vw),12px)}
js
let aligned=false;function draw(){const xs=aligned?[46,52,56]:[20,50,80];['m1','m2','m3'].forEach((id,i)=>document.querySelector('#'+id).style.left=xs[i]+'%');document.querySelector('#band').style.left=xs[0]+'%';document.querySelector('#band').style.width=(xs[2]-xs[0])+'%';document.querySelector('#spread').textContent=aligned?'SPREAD: NARROW':'SPREAD: WIDE'}draw();setInterval(()=>{aligned=!aligned;draw()},1750);

Attribution allocates credit along observed paths, MMM models aggregate spend and outcomes, and lift tests compare a counterfactual. They answer related but different questions over different data.

The demo places three illustrative estimates for one channel on a common axis and highlights their spread. The goal is to investigate disagreement, not force identical numbers.

Align periods, audiences, and conversion definitions before comparing. Record uncertainty and bias for each method.

When to use

Use it before major budget changes to compare the direction of independent evidence.

Open as page ↗