Latest / The Psychology Undergrad Podcast / Applied Behavior Analysis & Single-Subject Research Designs (ABAB & Multiple-Baseline)
Transcript
- 0:00Welcome to the Psychology undergrad. Hey, everyone.
- 0:02So glad you're here. Yeah, we are so glad you
- 0:05could join us for this study session today. Grab
- 0:07a seat, maybe a coffee, because today our deep
- 0:11dive into the Chapter 12 source material. is
- 0:15taking us straight into something super fascinating.
- 0:18Oh, totally. We're getting into the mechanics
- 0:19of single -subject research designs. Right. Which
- 0:22sounds dry, but I promise it's actually mind
- 0:24-blowing. It's basically asking, what if you
- 0:26could scientifically prove that a psychological
- 0:28intervention works without needing, you know,
- 0:32massive lecture halls full of hundreds of participants?
- 0:35Yeah, or like... complex statistical algorithms
- 0:38that take a supercomputer to run. Exactly. What
- 0:41if the only data point you ever needed to change
- 0:43a behavior was, well, you? It is such a radical
- 0:46departure from how we usually picture psychology.
- 0:49I mean, we spend so much time in this field looking
- 0:51for the average response across these huge populations,
- 0:54right? Oh, for sure. Everything is about the
- 0:56mean, the standard deviation, the group average.
- 0:59But the methodologies we are dissecting today,
- 1:02they allow researchers to definitively prove
- 1:05cause and effect using just one single individual.
- 1:09Which is why this specific topic feels so intensely
- 1:12personal to me. I mean... Think about your own
- 1:15life for a second, right? We are constantly running
- 1:17these sloppy little experiments on ourselves.
- 1:19Oh, absolutely. All the time. You try a new study
- 1:22habit, hoping it bumps up your GPA, or I don't
- 1:25know, you test out some bizarre morning routine
- 1:27because a TikTok influencer promised it would
- 1:29maximize your productivity. Ice beds at 5 a .m.
- 1:32Yes, exactly. But when we see a change, we are
- 1:36usually just guessing about what actually caused
- 1:39it. Like, did the ice bath work, or did I just
- 1:41finally get eight hours of sleep? Right. There
- 1:43are way too many variables. Exactly. So mastering
- 1:46how to measure individual change with actual
- 1:48scientific rigor is like the ultimate life hack.
- 1:51You stop guessing, and you become your own hyper
- 1:54-accurate science experiment. To reach that level
- 1:57of accuracy, though, we have to follow a pretty
- 1:59rigid set of experimental rules. You know, you
- 2:02can't just adopt a new morning routine, notice
- 2:04you feel a bit more energized on a Tuesday, and
- 2:07declare the routine a scientific success. Yeah,
- 2:09sadly, vibes don't count as science. They really
- 2:12don't. Coincidences happen every single day.
- 2:15To prove that your specific intervention caused
- 2:18the change, you have to establish a controlled
- 2:21environment within your own daily life. Okay,
- 2:24so let's figure out how to build that controlled
- 2:25environment because before we introduce any new
- 2:28habit or, you know, treatment, we need a starting
- 2:30line. Right, the baseline. Yeah, and in these
- 2:33single subject designs, the study is divided
- 2:35into phases and they label them with capital
- 2:37letters. So phase A is always your baseline,
- 2:40basically your life before the intervention.
- 2:43And then phase B is your treatment condition.
- 2:46But researchers don't just arbitrarily measure
- 2:48the baseline for a week and then like blindfold
- 2:51themselves and jump into phase B. No, they have
- 2:53to be way more methodical than that. Exactly.
- 2:56They rely on this foundational principle called
- 2:58the steady state strategy. Sidman outlined this
- 3:01way back in 1960, I think. Yeah, 1960. Right.
- 3:05So a researcher will continuously measure the
- 3:07target behavior during that baseline phase A,
- 3:10and they will stubbornly refuse to introduce
- 3:12the treatment until that behavior becomes consistent.
- 3:15like from one observation to the next. It has
- 3:17to be stable. The best way I found to visualize
- 3:20why the steady state strategy is so critical
- 3:22is to think of your daily behavior like a radio
- 3:25signal. Oh, I like that analogy. Yeah, so if
- 3:28there is a ton of static on the station, maybe
- 3:31you are super stressed about midterms, your sleep
- 3:34schedule is completely wrecked, your caffeine
- 3:36intake is all over the map, you literally can't
- 3:38hear the music. Right, that static is what researchers
- 3:41call noise in the data. Exactly, noise. So if
- 3:44you try to introduce a new study technique while
- 3:46your life is producing that much noise and your
- 3:48grades go up slightly, you have no way of knowing
- 3:50if the technique worked or if you just happen
- 3:53to, you know, drink less coffee and sleep more
- 3:56before the exam. Yeah, the steady state strategy
- 3:58is essentially turning down the static until
- 4:00you get a clear, consistent tone. Only then do
- 4:03you introduce the treatment to see if it actually
- 4:05changes the tune. That reduction of noise is
- 4:08the absolute prerequisite for proving causation
- 4:11in a single subject. Like, if your baseline behavior
- 4:14is stable, meaning the data points are just hovering
- 4:17around a predictable level without wildly spiking
- 4:20or crashing, then any dramatic shift that occurs
- 4:23the exact moment you end for phase B becomes
- 4:26highly suspicious. Suspicious in a good way,
- 4:28right? Yeah, exactly, in a good way. The steady
- 4:30state ensures that the effect of your independent
- 4:32variable isn't masked by random daily chaos.
- 4:36Okay, but minimizing the static and getting that
- 4:38flat baseline, it doesn't actually prove our
- 4:40new study habit works, it just gives us a clean
- 4:43slate. True. To prove cause and effect beyond
- 4:45a shadow of a doubt, we have to intentionally
- 4:47disrupt that steady state. Which brings us to
- 4:50the most fundamental framework in this whole
- 4:53chapter, the reversal design. Also known as the
- 4:55ABA design. It operates on this. brutal, but
- 4:59really effective logic. Brutal is a good word
- 5:00for it. Yeah, because we start with phase A,
- 5:03measuring the baseline until we achieve that
- 5:05steady state. Then we introduce our intervention
- 5:08in phase B. We monitor the behavior until it
- 5:11stabilizes at a new, hopefully improved, steady
- 5:15state. And here is the twist. To prove the intervention
- 5:18is what actually caused the improvement, you
- 5:21don't just celebrate and go home. You enter the
- 5:23second phase A. You actively remove the treatment.
- 5:26Which is wild. You take it away and wait for
- 5:29the behavior to return to its original baseline
- 5:31steady state. There's a classic demonstration
- 5:33of this by Vance Hall and his colleagues. They
- 5:36wanted to increase the amount of time a student
- 5:38named Robbie spent actually studying in class.
- 5:41Classic Robbie. Always distracted. Right. So
- 5:43they tracked his baseline studying behavior,
- 5:46which was predictably dismal. Once it established
- 5:48a low steady state, they introduced the Phase
- 5:51B treatment. This involved the teacher providing
- 5:53highly specific positive attention whenever Robbie
- 5:56was actually on task. And let me guess, his studying
- 5:59time skyrocketed. It did, yeah. It skyrocketed
- 6:02and leveled off at a high steady state. But to
- 6:05prove it was the positive attention, they had
- 6:06to take it away. Oh, man. Poor Robbie. I know.
- 6:10They literally instructed the teacher to stop
- 6:12giving Robbie that specific feedback. And, predictably,
- 6:17his studying crashed right back down to his baseline
- 6:19levels. Wow. Finally, to make it a full AB AB
- 6:23design, they reintroduced the positive attention,
- 6:26and his studying shot right back up. It's crazy.
- 6:29It acts just like a light switch. Flipping it
- 6:31on, off, and on again proves that the switch
- 6:33is actually what controls the light. Exactly.
- 6:36And researchers can get creative with this. They
- 6:37sometimes use an ABC -ACB design, which allows
- 6:41them to test multiple treatments. Oh, right.
- 6:43Because phase B might be positive attention,
- 6:45while phase C could be, I don't know, a mild
- 6:47verbal reprimand for distraction. Yeah. And by
- 6:50alternating and reversing these phases, you can
- 6:52pinpoint exactly which intervention works best
- 6:55for that specific individual. OK. I understand
- 6:57the mechanism, but let me play devil's advocate
- 6:59for a second here. Go for it. If I measure my
- 7:01baseline studying, start a new focus technique,
- 7:03and my grades shoot up and establish a new high
- 7:06steady state, why isn't that just an A -B design?
- 7:10Like, why do I have to purposefully break my
- 7:13good habits just to prove they worked? That's
- 7:15a great question. It's because an A -B design
- 7:18is fundamentally just an interrupted time series.
- 7:21It severely lacks internal validity. Meaning
- 7:23we don't actually know if the treatment caused
- 7:26the change? Right. Think about the vulnerabilities
- 7:28of that setup. If your grades improve the exact
- 7:31week you start your new focus technique, how
- 7:34do you know the technique caused the improvement?
- 7:36What if your professor decided to start grading
- 7:39on a super generous curve that exact same week?
- 7:42Or what if your noisy roommate moved out and
- 7:45your apartment suddenly got much quieter? Oh,
- 7:47I see. So an extraneous variable snuck in and
- 7:50basically hijacked the results. Precisely. If
- 7:52you only look at phase A to phase B, you cannot
- 7:54rule out that an extraneous variable just happened
- 7:57to perfectly coincide with your new habit. But
- 7:59if you remove the habit and your grades immediately
- 8:01drop, and then you restart the habit and they
- 8:03immediately rise, the mathematical probability
- 8:06of an extraneous variable perfectly syncing with
- 8:09your on and off schedule three or four times
- 8:11in a row becomes practically zero. The reversal
- 8:15is the definitive proof. Exactly. OK, the math
- 8:17makes total sense. Scientifically, taking the
- 8:20treatment away is a brilliant safeguard. but
- 8:23ethically. Yeah, this is where it gets heavy.
- 8:25It really does. This brings us to a massive glaring
- 8:29moral dilemma. It might honestly be one of the
- 8:32most profound conflicts of conscience in psychological
- 8:36research. Absolutely. Because the very mechanism
- 8:38that gives the ADA design its scientific rigor
- 8:41is exactly what can make it deeply profoundly
- 8:44unethical to use. We see this most acutely in
- 8:47clinical settings, especially with vulnerable
- 8:49populations. Imagine a researcher working with
- 8:52a developmentally disabled child who engages
- 8:55in severe self -injurious behavior. That's so
- 8:58tough. It is. The researcher designs an intervention
- 9:01that successfully stops the child from hurting
- 9:03themselves. The behavior reaches a steady safe
- 9:06state in phase B. But then, to satisfy the scientific
- 9:10requirements of the reversal design, that researcher
- 9:13is supposed to deliberately withdraw the protective
- 9:15treatment. Yes. They have to withdraw it and
- 9:18watch to see if the child begins injuring themselves
- 9:21again. That goes against every moral instinct
- 9:23we have as humans. It really does. It touches
- 9:26on broader principles of stewardship, compassion,
- 9:29and just our fundamental duty to protect the
- 9:32vulnerable. you are actively endangering a human
- 9:35being just to satisfy the requirements of a line
- 9:38graph. It creates a direct collision between
- 9:40the pursuit of pure objective scientific knowledge
- 9:43and the fundamental medical imperative to do
- 9:45no harm. Yeah, you are forcing the clinician
- 9:47to ask if the statistical proof of the cure is
- 9:50more valuable than the actual safety of the patient.
- 9:53And in any ethical framework, the patient's well
- 9:56-being must absolutely override the experimental
- 9:58design. Without a doubt. And beyond the heavy
- 10:01moral implications, there's also just a massive
- 10:03practical flaw with the reversal design. Oh,
- 10:05you mean when the treatment works too well? Yes.
- 10:08What if the treatment is so good that it creates
- 10:10a permanent change? Like, if a therapist uses
- 10:13a cognitive intervention that completely cures
- 10:16a patient's phobia, removing the treatment won't
- 10:19cause the phobia to return. Right, because the
- 10:21patient learned a whole new way of thinking.
- 10:23Exactly. The behavior won't reverse to baseline.
- 10:26And ironically, under the strict rules of an
- 10:29ABA design, because the behavior didn't reverse,
- 10:31you technically failed to prove your intervention
- 10:33worked. It's a fascinating paradox. The treatment
- 10:37was so flawlessly effective that it completely
- 10:39broke the scientific tool used to measure it.
- 10:42Because of these moral failures and, you know,
- 10:45practical paradoxes, researchers desperately
- 10:47needed a new methodology. They needed a way to
- 10:50prove cause and effect with the exact same mathematical
- 10:53certainty as a reversal design. but without ever
- 10:56having to strip away a successful treatment.
- 10:59And this necessity birthed the multiple baseline
- 11:01design. Yes. The multiple baseline design sidesteps
- 11:04the ethical dilemma entirely by utilizing staggered
- 11:08introduction rather than removal. Staggered introduction.
- 11:11So instead of turning the treatment on and off
- 11:13for one scenario, you stagger the rollout of
- 11:16the treatment across different scenarios. Exactly.
- 11:19And researchers typically execute this in three
- 11:21distinct ways. OK. To keep these straight, I
- 11:24like to use a simple who, what, where mental
- 11:27framework. That's a great way to remember it.
- 11:29So the who represents multiple baselines across
- 11:32participants. The what represents multiple baselines
- 11:35across different behaviors or variables. And
- 11:38the where represents multiple baselines across
- 11:41different settings. Let's unpack the who first.
- 11:44In a multiple baseline design across participants,
- 11:46you establish a steady state baseline for several
- 11:49different individuals at the exact same time.
- 11:52OK. Crucially, you do not apply the treatment
- 11:54to all of them at once. You introduce it to participant
- 11:56one, wait, then introduce to participant two,
- 11:59wait, and so on. Ah, I remember this. Scott Ross
- 12:02and Robert Horner published a fantastic study
- 12:04using this exact method back in 2009. The bullying
- 12:08study, right? Yes. They were testing a school
- 12:11-wide bullying prevention program. They selected
- 12:14three different elementary schools. And at each
- 12:17school, they pinpointed two specific students
- 12:20who had severe histories of physical and verbal
- 12:22aggression. And the precision of their observation
- 12:25was remarkable. During the baseline phase, observers
- 12:29monitored these specific children during lunch
- 12:31recess every single day. Right. And they used
- 12:33those handheld computers, didn't they? Yeah.
- 12:35They used handheld computers to log the exact
- 12:38frequency of aggressive acts within 10 -minute
- 12:41windows. That established a very clear, quantifiable,
- 12:45steady state of bad behavior for all three schools
- 12:48simultaneously. So once that baseline was rock
- 12:50solid across the board, they rolled out the anti
- 12:53-bullying program, but only at School 1. Just
- 12:55School 1. And the aggressive behavior at School
- 12:571 plummeted. Now, if they had rolled it out to
- 13:00all three schools on the same day and aggression
- 13:02dropped everywhere, a skeptic could easily dismiss
- 13:05it, right? Oh, for sure. They could argue that
- 13:07the weather suddenly got too hot for kids to
- 13:09run around fighting or, I don't know, that a
- 13:13highly anticipated video game was released and
- 13:15all the kids were just talking about that instead
- 13:16of bullying each other. But Ross and Horner staggered
- 13:19the rollout. They waited two full weeks before
- 13:22introducing the program to School 2. An aggression
- 13:24at School 2 remained high during those two weeks,
- 13:27proving the weather or a video game hadn't changed
- 13:30anything. The drop only occurred the moment the
- 13:33program was introduced. They waited another two
- 13:35weeks before implementing it at School 3, and
- 13:38again, the behavior only changed precisely when
- 13:40the intervention arrived. The mathematical absurdity
- 13:43of a coincidence makes this design so bulletproof.
- 13:47I mean, an extraneous variable like a weather
- 13:50change would have had to perfectly align with
- 13:52their staggered two -week schedule three completely
- 13:54separate times. Which is basically impossible.
- 13:56Right. It proves the intervention works not by
- 13:58taking it away, but by proving it works every
- 14:01single time you turn it on in a new environment.
- 14:04It's brilliant. Okay, so moving to the... What
- 14:09in our framework? Multiple baselines across behaviors.
- 14:13This involves a single participant, but you track
- 14:16several independent behaviors. Yeah, imagine
- 14:19an employee struggling with productivity. You
- 14:21might establish a baseline for how many client
- 14:23calls they make. and a totally separate baseline
- 14:26for how many reports they draft. So you introduce
- 14:28a time management intervention, but you strictly
- 14:31apply it to the client calls. You completely
- 14:33ignore the report drafting. Exactly. If the calls
- 14:36increase while the reports remain stagnant, you
- 14:38have isolated the treatments effect. And then
- 14:41a few weeks later, you apply the same intervention
- 14:43to the report drafting. When that metric jumps,
- 14:46you've confirmed the internal validity without
- 14:48ever removing the helpful time management tool.
- 14:50You've got it. Finally, there's the rain. were
- 14:53multiple baselines across settings. You track
- 14:55one specific behavior in one specific person,
- 14:59but across different environments. So, like,
- 15:01you might measure a child's disruptive outbursts
- 15:03at home, in the classroom, and, I don't know,
- 15:05at the grocery store. Right. You teach the parents
- 15:08an intervention to use at home first. Once the
- 15:11outbursts at home drop and establish a new steady
- 15:14state, you teach the intervention to the classroom
- 15:16teacher. It beautifully fulfills the moral duty
- 15:19of care. You never have to take away a treatment
- 15:22that is helping a child, yet you maintain absolute
- 15:24scientific rigor by staggering the wear. It illustrates
- 15:28how scientific methodology can beautifully adapt
- 15:30to honor ethical standards without sacrificing
- 15:33the validity of the data. OK, so we have successfully
- 15:37gathered all this data ethically. We established
- 15:39our steady states. We minimized our noise. And
- 15:42we staggered our baselines perfectly. But this
- 15:45leads us to a pretty fascinating divergence from
- 15:47traditional psychology. The grading part. Yes.
- 15:50In massive group studies, researchers feed all
- 15:53their data into statistical software to hunt
- 15:55for microscopic p -values hidden in the averages.
- 15:59Single -subject research, however, relies almost
- 16:01entirely on visual inspection. Right. They literally
- 16:03plot the data. on a graph and use their own eyes
- 16:07to determine if the treatment worked. Which,
- 16:09I have to admit, I initially struggled with the
- 16:11idea of visual inspection. It sounds like, you
- 16:13know, vibes -based science. Just eyeballing a
- 16:15graph feels highly subjective, yeah. But looking
- 16:18at these criteria, it actually demands an overwhelming
- 16:21amount of proof. Let me see if I can break down
- 16:23the first one. Level. Go for it. This is basically
- 16:26the raw altitude of the data points, isn't it?
- 16:29If the baseline behavior is hovering around 10
- 16:32occurrences a day, and the moment you start treatment,
- 16:34the data points cluster around 50 occurrences
- 16:37a day, that massive jump in the axis is your
- 16:40change in level. That is exactly it. A clear,
- 16:43undeniable shift in the average magnitude of
- 16:45the behavior. But level alone isn't enough. Which
- 16:49brings us to trend. Trend focuses on the slope
- 16:51of the data points over time, right? Yeah. Are
- 16:54the points generally creeping upward or sliding
- 16:56downward? This becomes incredibly persuasive
- 16:59when the intervention intercepts and reverses
- 17:01the momentum. Oh, right. So if a harmful behavior
- 17:04was steadily climbing steeper and steeper during
- 17:06the baseline phase, and the introduction of the
- 17:08treatment causes it to immediately begin slipping
- 17:10downward... That sharp V -shape on the graph
- 17:13is a powerful visual argument for causation.
- 17:16Okay. And the third one is latency, which is
- 17:18all about speed, right? It measures the delay
- 17:21between the moment you introduce the treatment
- 17:23and the moment you see a change in the level
- 17:25or the trend. It is the reaction time of the
- 17:27behavior. If you introduce a new study habit
- 17:30on Monday and your focus immediately sharpens
- 17:33on Monday afternoon, the latency is practically
- 17:35zero. And that immediate reaction strongly implies
- 17:38the habit was the catalyst. Right. If, however,
- 17:41it takes four weeks for your focus to slowly
- 17:44begin improving, That long latency leaves the
- 17:47door wide open for extraneous variables. A lot
- 17:50of other things in your life could have changed
- 17:52over those four weeks to cause the improvement.
- 17:54So when a graph shows a massive jump in level,
- 17:57a sharp reversal in trend, and an immediate latency,
- 18:00the effect hits you over the head. You don't
- 18:02need a complex computer algorithm to tell you
- 18:04it worked. No, the visual evidence is undeniable.
- 18:08But I know researchers still crave some objective
- 18:10math to back up their eyes sometimes. Which brings
- 18:12us to PND, or the percentage of non -overlapping
- 18:15data. Scruggs and Master Pieri championed this
- 18:17in 2001. They sure did. PND is an elegant bridge
- 18:22between visual inspection and statistical quantification.
- 18:26It's specifically designed to handle the reality
- 18:28that baselines are rarely perfectly flat lines.
- 18:32They usually have some natural bounds. The way
- 18:34you calculate PND is wonderfully strict. You
- 18:37look at your entire baseline phase and you find
- 18:40the single highest outlier. like the absolute
- 18:43best day you had before the treatment started.
- 18:45Yep, the peak. Then you look at your treatment
- 18:48phase and you calculate what percentage of those
- 18:50new data points are higher than that one baseline
- 18:53peak. It forces your treatment to beat your absolute
- 18:56best pre -treatment day over and over again.
- 18:59If we look back at the study on Robbie's studying
- 19:02behavior, his PND was 100%. Wow, 100%. Yeah.
- 19:06Every single time they measured his studying
- 19:08during the positive attention phase, the duration
- 19:10was higher than his absolute highest recorded
- 19:12spike during the baseline phase. That is insane.
- 19:14So the higher the percentage of non -overliving
- 19:17data, the more robust the intervention. It quantifies
- 19:19the visual impact without requiring inferential
- 19:22statistics? Exactly. Man, this is so cool. Let's
- 19:25pull all these threads together for our final
- 19:26review. Sounds good. If you want to scientifically
- 19:29measure change in a single subject, you must
- 19:32start with a steady state. You have to turn down
- 19:35the radiostatic before you try to change the
- 19:38station. Right. And then we explore the ABA reversal
- 19:40design, which secures internal validity by turning
- 19:43the treatment on and off. But we also grappled
- 19:46with the heavy moral baggage of intentionally
- 19:49removing a successful intervention from someone
- 19:51who desperately needs it. Which logically led
- 19:54us to the multiple baseline solution. It preserves
- 19:56the ethical duty of care by staggering the introduction
- 20:00of the treatment across different participants,
- 20:02different behaviors, or different settings. Our
- 20:04who, what, where framework. And finally, we evaluate
- 20:07all the staggered data, not with group averages,
- 20:10but through rigorous visual inspection. We're
- 20:13judging the level, the slope of the trend, the
- 20:15speed of the latency, and verifying it all with
- 20:18the strict math of the percentage of non -overlapping
- 20:21data. It requires a profound shift in perspective.
- 20:25You know, you stop asking what works for the
- 20:27hypothetical average person, and you demand proof
- 20:30of what definitively works for the unique individual
- 20:33sitting right in front of you. Which leaves us
- 20:35with a provocative final thought for this study
- 20:37session. If these designs are how we prove what
- 20:40works for an individual, maybe it's time we start
- 20:42applying them to ourselves. Ooh, I love that.
- 20:45I mean, we devour self -help podcasts and books
- 20:47blindly trusting that what worked for a group
- 20:50of strangers in some university study will magically
- 20:53work for us. But you aren't an average. No. You
- 20:55have a unique brain, unique stressors, and a
- 20:58unique environment. Exactly. So what if, instead
- 21:00of guessing, you treated your next personal goal
- 21:02as a strict, single -subject design? Measure
- 21:05your baseline. Tolerate the noise until you find
- 21:08your study state. Introduce one variable. Grade
- 21:11your own level, trend, and latency. It might
- 21:14be time to realize that in the pursuit of optimizing
- 21:17your own life, you are perfectly equipped to
- 21:19be both the lead scientist and the sole subject.
- 21:22The most reliable data you will ever gather is
- 21:25your own. Well said. Let that concept marinate
- 21:27for a while. Thanks for diving deep into the
- 21:30material with us today on The Psychology Undergrad.
- 21:33We'll see you at the next study session.