You set up your intervention. The behavior changes. Which means great — except was it your intervention, or just the slow march of Tuesday turning into Friday? That's the headache when using a multiple baseline across behaviors design in single-case research. You're trying to prove cause and effect without ever withdrawing treatment, and that's harder than it sounds.
It sounds simple, but the gap is usually here And that's really what it comes down to..
Most people hear "multiple baseline" and picture something tidy. It's a careful, slightly awkward dance where you stagger your start times across different behaviors in the same person, or across the same behavior in different people, or even different settings. Day to day, it isn't. And the version we're digging into here is the one where you target multiple behaviors in one subject No workaround needed..
Real talk — this step gets skipped all the time.
What Is a Multiple Baseline Across Behaviors Design
Look, the short version is this: you pick two or three behaviors that don't influence each other, measure them all at once, and then introduce your intervention to one behavior at a time. The others keep getting measured but stay untouched — at least until you've shown the first one moved Turns out it matters..
Here's the thing — it's not a wait-and-see approach by accident. It's by design. You're banking on the idea that if Behavior A changes the moment you intervene, but Behaviors B and C sit flat, then A's change is probably yours. In practice, then you intervene on B. If B moves and C doesn't, you've got another data point. By the time C finally gets the treatment and also shifts, the pattern is doing the arguing for you That alone is useful..
Why "Across Behaviors" Specifically
So why not just do one behavior and call it a day? Because some behaviors can't be reversed. You can't un-teach a kid to read to prove your reading program worked. And you shouldn't withdraw a self-injury intervention just to satisfy a research design. Multiple baseline across behaviors lets you keep the help on, while still building a case.
The "across behaviors" part means the staggered baselines are different responses from the same individual. Maybe it's a child who has three separate problem behaviors: tantrums, noncompliance, and interrupting. Or an adult in therapy working on anxiety, sleep, and avoidance. That's why same person. Different targets.
This changes depending on context. Keep that in mind.
What Makes a Behavior a Good Candidate
Not every behavior plays nice in this design. In practice, you need targets that are independent. Which means if you fix tantrums and suddenly noncompliance drops because they're the same underlying thing, your design falls apart. You also want behaviors you can measure reliably — daily, ideally. And they should matter. Don't run a multiple baseline on nail-biting, pencil-tapping, and humming unless those are actually the problem Surprisingly effective..
Why It Matters
Why does this matter? Because most real-world interventions can't be turned off. In clinics, classrooms, and homes, you don't get to do a clean ABAB reversal. Think about it: ethics stop you. Reality stops you. A multiple baseline across behaviors design is one of the few tools that respects that and still gives you science The details matter here..
This is where a lot of people lose the thread.
Turns out, it also protects you from the biggest threat in single-case work: the coincidental improvement. Worth adding: the dog dies. Mom gets a new job. Life happens. Because of that, a new teacher shows up. Practically speaking, any of those could shift behavior. But if three behaviors are tracked and only the treated one moves at the right moment — twice, then three times — the dog theory loses steam That's the part that actually makes a difference..
And here's what most people miss: this design isn't just for publishing papers. Teachers use it to show a principal the new token system works. Parents use it to see if a bedtime routine actually helped. It's practical evidence, not just academic jewelry That's the part that actually makes a difference..
How It Works
The meaty part. Let's walk through it like you're actually setting one up.
Step 1: Pick Your Behaviors and Define Them
You need clear operational definitions. "Says 'no' to adult requests within 10 seconds, three or more times per hour" is. "Bad attitude" isn't a behavior. On the flip side, have someone else watch a video and see if they tag the same thing. So write these down. That's interobserver agreement, and without it your whole design is a house of cards.
You're looking for three behaviors, usually. Plus, two can work, but three gives a stronger demonstration. They must be functionally separate. If you're not sure, pilot it. Watch for a week.
Step 2: Establish Concurrent Baselines
Here's where the "multiple" starts. You measure all behaviors at the same time, every day, under the same conditions. And no intervention yet. You're waiting for stability — not flat lines, but a predictable pattern. A rising trend is stable if it keeps rising the same way.
In practice, this is the boring part. Think about it: you're collecting data and resisting the urge to jump in. Most people can't wait. They see a behavior they hate and want to fix it now. Don't. Worth adding: the baseline is the control condition. Short it, and you've shorted the whole study Which is the point..
Step 3: Intervene on the First Behavior
Pick one. And usually the most disruptive or the one the person cares about most. Apply your intervention. Keep measuring the others with zero change to their conditions The details matter here. Still holds up..
Now watch. Even so, if the first behavior changes and the others don't, that's your first clue. Now, the length of baseline before intervention should differ across behaviors — that's the stagger. Which means maybe Behavior A gets intervened at day 10, B at day 18, C at day 25. The different start points are what let you rule out "everything got better because spring came The details matter here. That alone is useful..
Step 4: Cascade the Intervention
Once A has shown a clear change and held it, you start the intervention on B. A keeps getting the intervention. That's why c is still baseline. If B shifts and C doesn't, good. Then C gets it. If C moves too, you've got a replication across behaviors.
The logic is simple but brutal: if all behaviors changed at the same time, your intervention timing didn't explain it. If they changed one by one, right when you touched each one, you're looking at a functional relation.
Step 5: Graph It Like Your Credibility Depends on It
Because it does. Standard convention: one panel per behavior, stacked vertically, same time axis. Because of that, anyone should be able to glance and see the stair-step pattern. Intervention lines dropped at different x-coordinates. If your graph needs a paragraph to explain, it's the wrong graph.
Common Mistakes
Honestly, this is the part most guides get wrong — they pretend the design is foolproof. It isn't.
One classic error: behaviors that aren't really independent. So you thought you had three baselines. Still, a kid learns to ask for breaks and tantrums drop — but so does noncompliance, because asking for a break was the compliant alternative. You had one behavior with three costumes.
Another: baseline too short. People intervene after four data points because that feels like enough. It isn't. Day to day, you need to see the pattern. A four-point baseline can look stable and then spike the day after you start, and now you can't tell if you caused it or the spike was coming anyway.
And the silent killer — inconsistent measurement. If your data on Behavior B is solid but Behavior C is "eh, looked about the same," the whole demonstration weakens. You can't stagger what you didn't reliably count Worth knowing..
But the worst mistake? In practice, picking behaviors nobody cares about to make the design clean. I've seen studies on "organizing pencils" and "saying hello" just because they were easy to count. Real talk — if the behavior doesn't matter to the person's life, the design is a party trick.
Practical Tips
Here's what actually works when you're in the trenches.
Start with a behavior you're already mad about. The one that's costing the most isn't just ethically easy to prioritize — it keeps you funded in attention. Not joking. You'll keep measuring the boring ones if the first one matters.
Use the same measurement system for all behaviors. Different tools per target invites error and makes the graph look like a ransom note. Same observation window, same recorder, same definition style.
Don't tell the participant "now we're working on B.So in practice, with kids, they usually figure it out. Here's the thing — " Okay, sometimes you must. But if you can keep the staggered start quiet, you avoid expectancy effects. With adults, be honest but don't make a ceremony of it.
Not obvious, but once you see it — you'll see it everywhere.
Plan your stagger before you start. Day 10, 18, 25 — write it down. If you decide "once it
looks stable" in the moment, you'll drift, and drift kills the logic of the design The details matter here..
One more thing that saves people: pre-write your rationale for each start date. Now, when the funding reviewer or your supervisor asks why you began the intervention on Behavior C on Day 25 and not Day 20, you should have the baseline trend in front of you, not a shrug. "It was flat for eight sessions" beats "felt right" every time Easy to understand, harder to ignore. Less friction, more output..
And if a behavior changes before its scheduled intervention — say, Behavior C improves during Baseline B's intervention — note it, don't panic. That's a spillover effect, and it's data, not disaster. You adjust the story you're telling about independence. Practically speaking, you report it. You don't pretend it didn't happen.
Conclusion
The multiple baseline design isn't magic, and it isn't clean in the way textbooks pretend. In practice, it's a disciplined bet: that by staggering when you act, you'll expose cause without having to withdraw support from anyone. It works when the behaviors are real, the measurement is honest, and the staggers are planned like you mean it. Get those three right, and the stair-step graph isn't a flex — it's proof. Get them wrong, and you've got a complicated way to waste paper. So pick what matters, count it like it counts, and let the timeline do the talking Less friction, more output..
