Next up was Michele Lobo who presented The New Face of Single Case Research Design: P Values, Randomization and Causality, Oh My!
The almighty randomized controlled trial (RCT), the epitome of research perfection…. or is it? I don’t know about you, but in school I was taught to pretty much ignore everything that wasn’t an RCT or systematic review (SR)– which takes a bunch of RCTs and runs their statistics together in 1 mega huge study. The RCT is the basically the backbone of the SR. Simply because there weren’t enough participants in the other studies to evenly distribute out other variables, that might confuse the results. For example, say you’re wanting to find the most attractive nose shape by showing faces on a screen and then asking participants to rate them 1-5 on an attractiveness scale. Well, I’m not much of a ginger fan (sorry, just not), so my ratings may be biased against gingers. If you have just 5 participants including me, you’d probably get a skewed result. BUT if you had 105 participants, my bias wouldn’t matter so much in the long run. That’s why everyone pushes the RCT or SR. However, RCTs do have some set backs. In therapy, there can be A LOT of confounding variables, so the randomization of the RCT may not be controlling variables as much as we think. It is difficult for the clinician researcher to gather enough individuals to participate in a study of that size. RCTs gather and report data on “the average.” Have you ever met a real life person that is that “average.” I don’t think so. They always have very strict inclusion and exclusion criteria, so again, they’re skewing the data away from the real person.
Enter the Single-Case research design. In this kind of study, there is one participant who serves as their own control. How can you get a control v. experimental group that is more closely related to oneself. Hint: you can’t, all variables, seen and unseen are controlled for exactly. Then you repeatedly observe one manipulated intervention over time. For example, let’s say you’re wanting to know how Drug X effects Condition A. You recruit a person with Condition A. You take whatever measurements are of most importance in Condition A when the person is not taking Drug X. Then you start the participant on Drug X at 5 mg for a couple weeks, take your measurements again, then stop the drug and let it wash out their system for whatever time is required. Then you start them up on Drug X again at 10mg… etc. Until you look at all the intensity levels you want. Now imagine that Drug X is a therapeutic intervention such as a procedural intervention, educational intervention, task modification, environment modification, feedback/reinforcement, assistive tech, level of assist…. You get the idea.
Single-Case is NOT
— a case study/case series
— lacking in internal or external validity (CAN determine causal relationship with ABAB design– baseline-> intervention -> baseline -> intervention– or AB design with multiple participants, CAN be generalized)
— lacking random assignment– have more than 1 participant, have at least 3 data points (recommend 5+), then randomize the participants to multiple conditions. This makes for lots of data (making a potentially strong effect), and requires computer program to analyze. More on that in a minute.
— only analyzed with visual analysis (level, variability of performance, direction/degree of trends) or descriptive statistics (mean, median, mode), but also interferential stats (parametric t-test, ANOVA, Mann-Whitney U, Kruskall-Wallis test– which I’ve never heard of, honestly)
— necessarily 1 participant (CAN have 1-infinity number of participants)
The Single-Case is particularly useful when there are individual variations in the effect expected; you’ve looking at a small population, like say with ALS; it may not be safe to not provide the intervention, like it is a life saving/sustaining treatment; you’re testing something completely new or in a new population; or if the intervention is very expensive. It also has many advantages over RCTs such as, being more responsive to needs of participants; you don’t need as many participants or resources (aka finances) as an RCT; isn’t looking at an “average” non-existent person and so reveals a real human being’s reaction to the treatment which is more natural to the manner in which the intervention would actually be applied clinically; and as discussed above, is generalizable and usable with stats to show significance and size effects.
But like anything else, it does have drawbacks. A researcher needs appropriate training/assistance to design and analyze, just like any other research design. Generalizability is limited until you increase the number of participants in the study. And since my opinion of everything outside RCTs and SRs is fairly common, not too many people pay attention to these studies– it is truly the red-headed step child of research design. And it isn’t the best design for every research question either, just like the RCT isn’t either. So from here, we need to work on changing the funding, scientific, and educational culture to be more accepting of other research designs, to improve evaluation of which design is right for which research question, to more objectively analyze published research and not throw the proverbial baby out with the bath water.
If you’re interested in learning more, the Institute for Education Sciences hosts training programs in the summer, including some free on-line courses. I like free! I’ll also put up Michele’s references, as she suggested reading through several of those books and said that info on R packages (software to do the analysis) was also included in those resources.
Bless anyone that does research on the ground. It mostly confuses me. But I think I might could do this kind. It makes sense clinically. After all, this is basically how we develop our clinical decision making skills/instinct. We don’t take specific data points on everything, but we can see patterns easily given enough patients. Now, running stats…. definitely would need to contract someone to help with that. But that’s the beauty of research, and the Single-Case design in particular: it doesn’t occur in a bubble.