 Hey everyone, welcome back. It's Veronica Howard. So we've talked about divided data. We've talked about stable data. Now we have to determine from those two metrics, are we convinced, are we convinced that the data are strong? And then finally, are we confident that our treatment and only our treatment was the cause of the behavior change? So returning to the example that we had before with Annie. Remember that we have a client who's very young, 12 months. They are showing some early signs of autism. We want to find a good intervention that's going to work to increase their social skills. We want to make sure they're connected with people. We, in our last couple of videos, have looked at doctors A and B who had very small treatment effects. So their data were not divided. We've looked at doctors C and D, one of whom intervened before they should have because the behavior was improving on its own. So really, why were they there? And Dr. D who had a pretty big treatment effect size, but their effect was not permanent. The data were creeping back towards baseline. Both doctors C and D had not stable intervention. So let's check out, this is our last chance. Dr. E is our last chance. This is Dr. E's treatment results. Would you hire Dr. E? So analyzing really quickly, remember on the Y axis or the left axis, Y is high. That's our dependent measure, the number of smiles. That's what we're interested in increasing on the X axis along the bottom. We have the number of sessions of behavior over time. And what I see here is that in the baseline condition before we start treatment, we had one smile, two smiles, zero smiles, two smiles, zero smiles. So not nearly enough smiles from our client. And then when we jump into the treatment condition, five smiles, seven smiles, nine smiles, eight smiles, nine smiles. So a big increase in the number of smiles. I would absolutely hire Dr. E. Look at these amazing differences. We see first of all, that there's no overlap between our baseline and our treatment condition. There's no hint, for instance, that the behavior is getting better on its own. There's no sign that the treatment is temporary. So in this particular case, we see this big change in the level of smiles, which is amazing, a huge dramatic increase. We see that these data are stable because they're not improving on their own. The treatment effects are not temporary. And we see such a big change. We go from an average of maybe one smile per session all the way up to maybe nine, eight or nine smiles per session, which is amazing. Of course, I would hire Dr. E. Now, when we're talking about convincing data, what we mean is we have data that are both divided and stable. So remember, data are, because data are plural, data are convincing if they're both divided and stable. If your data are divided, but they're not stable, or if your data are stable, but they're not divided, your data are not convincing. And you have to have both division and stability in the data to say that they're convincing data. But that's not enough, right? Convincing data, that's a good start. And you can demonstrate convincing data in a comparison design. We need to be able to show what's called a functional relation, a functional relation between your independent variable and your dependent variable or between your treatment and your behavior. You demonstrate that when you can show that your behavior changed when and only when. And I want to say this one more time because those four words are very important. Experimental control, a functional relation is demonstrated when and only when the behavior changes when treatment is put into place, but it doesn't change when the treatment is not put into place. So functional relation means that your behavior changes when and only when the treatment is put into place. So you have to have a strong design in order to determine if you have causal relation or a functional relation between your treatment and the behavior. You had to have convincing data and you have to have a strong experimental design. Now in this class Miller only talks about the comparison design, the reversal design, and the multiple baseline design. We've added on, we've talked about the changing criterion's design, which I think is a wonderful design, but comparison design is not enough, okay? Comparison design is very, very weak because any, any old thing can happen in there. The strong designs are reversal design and multiple baseline design. You can also add on to that changing criterion design. It's only with a reversal design, a multiple baseline design and a changing criterion design that you can demonstrate that your intervention and only your intervention was the cause of the change. You're ruling out confounding variables. Let me show you what I mean. So this is a very simple comparison design. Y axis shows me what the behavior of interest is and the X axis shows me the time sessions. So baseline levels one, two, zero, three, or excuse me, one, two, zero, two, zero, one, two, zero, two, zero. And then we get into treatment. We have five, seven, eight, nine, eight, nine, but that's only a comparison design. Anything, anything at all could have happened in the life of our client when we started treatment. Doesn't have to be just our treatment alone. So we can't say for sure that it's our treatment and not something else. If I want to be more confident that it's our treatment that's making that change, I'm going to reverse back to baseline conditions. I'm going to remove the independent variable if possible. And then if we see that the behavior also returns to baseline levels when it's under the same conditions, then I have more confidence in the functional relation between those variables. So we have baseline one, two, zero, two, zero, treatment five, seven, eight, nine, eight, nine. And then when we reverse to baseline, we see four, three, two, zero, one, zero. So that's a pretty big difference. So low data and baseline, high data and treatment, low data and baseline. Now if I'm still not convinced, I could put the treatment back in. And in this condition, we see that we increase the dependent measure to five, seven, eight, nine, 10. So the treatment goes back up there, the behavior goes back up when treatment is in effect. And we can put treatment back in to give us another example that the behavior changed when and only when the treatment was put into effect. Heck, if you're feeling like a mad scientist, you could just hokey pokey, just put treatment in, take treatment out, put treatment in, take treatment out. Now, obviously for practical reasons, you don't want to do that. It can be really disruptive to the life of the client. So don't do that. But you could to demonstrate experimental control if you were kind of a mad scientist. Now you can also demonstrate experimental control with multiple baseline design. This is, we use a multiple baseline design when it's impractical, impossible or unethical to remove treatment. So for instance, I'm showing an example here where this is a published study. It's a lovely study where we're teaching social skills using games, board games. And we have three different clients I'm showing you here, Billy, Carl and William. And we see that for these first two clients, Billy and Carl, the dependent measure is their spontaneous interactions with one another. We have three conditions, baseline training and post training, just to see if there's maintenance. We see that the behavior increases during the game training condition, and we show experimental control with the stagger. When we see for Billy and Carl that their behavior changes during game training, but remains unchanged for William and Andrew during their baseline, when they're still in baseline, but Carl and Billy have started treatment, that shows us behavior changes when and only when treatment begins. And we can even add more participants. So we are adding on Matt and Paul here, and we're showing that their behavior remains low during the baseline conditions, relatively unchanged and only increases during the treatment condition. Now, this is a very advanced multiple baseline design, but I just wanted to give you this example so you could see what we're talking about here. So just a quick summary here at the end. Remember that when we're talking about determining if we have convincing data and trying to figure out if we have a causal relation here, convincing just means that the data were divided and stable. Remember, I said data were because data are plural, and data are convincing if they're divided and stable. I can determine that my treatment caused the change in the behavior if I have convincing data, so they were divided and stable, and I have a strong design. The strong designs, the designs that can rule out confounding variables that can show me that my intervention and only my intervention was the cause of the change, those include reversal designs, multiple baseline designs, and changing criterion designs. It does not include a comparison design. Comparison designs are very, very weak. They cannot show you a functional relation. So strong designs have to be able to tell you that it was your treatment and only your treatment that caused the change in the behavior shows that by demonstrating behavior changes when and only when treatment is put into place, so you have to use reversal, multiple baseline, or changing criterion design. Let me know if you guys have any questions. Come on back for some practice materials, and I'll see you guys next time.