Geo lift experiment

지역 증분 실험

Run media in one of two comparable regions and compare outcomes.

···
html
<div class="geo"><div class="head">MATCHED REGIONS <b id="phase">PRE-PERIOD</b></div><div class="regions"><div class="region"><span>REGION A · MEDIA ON</span><div class="bar" id="a"></div><strong id="av">80</strong></div><div class="region"><span>REGION B · CONTROL</span><div class="bar ctrl" id="b"></div><strong id="bv">78</strong></div></div><div class="note" id="note">SIMILAR BEFORE TEST</div></div>
css
.geo{width:min(86%,590px);font:850 clamp(10px,max(2vmin,1vw),14px)/1.2 sans-serif}.head{display:flex;justify-content:space-between;border-bottom:2px solid var(--fg);padding-bottom:8px}.head b{color:var(--accent)}.regions{display:grid;gap:clamp(10px,3vmin,24px);margin:clamp(14px,4vmin,28px) 0}.region{display:grid;grid-template-columns:clamp(74px,18vmin,125px) 1fr clamp(22px,6vmin,40px);align-items:center;gap:8px}.region span{font-size:clamp(10px,max(1.6vmin,1vw),12px)}.bar{width:70%;height:clamp(18px,5vmin,34px);background:var(--accent);transition:width .75s}.bar.ctrl{background:var(--accent-2)}.region strong{text-align:right}.note{border-top:1px solid var(--line);padding-top:8px;color:var(--muted);letter-spacing:.08em;font-size:clamp(10px,max(1.7vmin,1vw),12px)}
js
let after=false;function draw(){document.querySelector('#phase').textContent=after?'AFTER MEDIA':'PRE-PERIOD';document.querySelector('#a').style.width=after?'93%':'70%';document.querySelector('#b').style.width=after?'73%':'68%';document.querySelector('#av').textContent=after?'112':'80';document.querySelector('#bv').textContent=after?'84':'78';document.querySelector('#note').textContent=after?'A MOVES ABOVE CONTROL':'SIMILAR BEFORE TEST'}draw();setInterval(()=>{after=!after;draw()},1800);

A geo lift experiment divides comparable regions into exposed and control groups to estimate media's additional effect. It can help when individual randomization is impractical or effects span channels.

In the demo, two regions track together before media starts, then one diverges. The target is that incremental difference, adjusted for pre-period behavior.

Check matching pre-trends and media spillover across regions. Few regions or major outside events increase uncertainty.

When to use

Use it to test cross-channel incremental sales at the regional level.

Open as page ↗