Recently I posed a question about research to fans of my Facebook page. The question goes like this:
Say you’re comparing dog-training method A for training a down-stay (using a clicker and food rewards) with method B (using a shock collar at low levels when the dog starts to get up). You want to know if one method causes more signs of fear and anxiety during the training than the other. For consistency you decide to use the same trainer for both groups. The trainer is skilled at both training methods. You are recording the training sessions and having a separate person observe the videotaped training sessions and record five separate measures of fear and anxiety (such as: yelping, cowering, ears back, tail down. You know which measures work well in this situation because you have already done a pilot study.
Now here comes the question.
Can this study be double blind? Can it be blind? Why or why not?
I received a medley of answers from Facebook fans and some of my student volunteers taking a stab at it. They ranged from:
It could be performed as a double blind. But it requires the tester to use a button that HE doesn’t know if it activates a clicker or a collar.
I think if you put the shock collar on both groups, but it was only used to elicit shock on 1 group it could be blind.
If the trainer administering the shock or click is removed from the presence of the dog and put, perhaps, in a room with a stimulus to administer an unknown signal (either a click or shock), they would be blinded from creating a bias in the experiment caused by unintentional visual cues or other cues from body language or tone of voice.
So what’s the answer?
To understand you first have to understand what blind and double blind mean.
Blind: In a blind study, either the subject or the observer (a.k.a. the researcher taking the measures) is unaware of which treatment each subject has received until after the experiment is over. In human research the first person “blinded” is generally the subject. But in animal behavior studies, it’s generally the observer who is blind.
Double blind: In a double blind study, neither the subject nor the person taking the measurements knows which treatment was received.
A Blind vs Double Blind Example
Lets start with an two examples. Lets say your questions is: Which cola tastes better, Coke® or Pepsi®. You bring test subjects (humans) into your lab and provide them with a sample of each. These people may already have a bias towards which brand they like and could therefore subconsciously choose their favorite based on recognizing the non-taste features such as the can. So in order to remove this bias you must “blind” them so they don’t know which one they are drinking. OK no, that doesn’t mean you have to put a blindfold on the test subjects. You can just present the drinks to them in identical looking cups so that they don’t know which product they are getting. Because the humans drinking the cola are unaware of which product they are sampling , the study is blind.
Now lets say we decide to look at something slightly different. We decide to look at the behavioral effects of drinking Coke® vs Pepsi® vs some super caffeinated new cola. The researchers decide they are going to measure change in blood glucose levels as well as having the subjects take a behavioral test of some sort.
They have each subject come in on 6 different Mondays. Each time, they get one of the three drinks. They are not be told which drinks they are sampling or what the tests are for because this may change the outcome. A technician will take their blood before and after they sample the drink and will administer the behavioral test once each session after the subjects have had their drink. The technician will not know which drink the subject has had because the drinks are just labeled A, B, and C. Only the researcher who has labeled the cups knows.
And when the blood sample is sent to the lab, the person running the sample also does not know which drink the person had. This experiment is double blind because neither the person taking the measure (the technicians), nor the subjects knows which experimental treatment they have received on a given day.
Why is it important that the subject not know? Well imagine this. You’re in a drug trial to test a possible wonder drug that helps anxiety. If you think you are going to be cured, the power of thought, which controls your body, may actually cause brain changes that improve your condition. So your improvement would not be due to the drug. It would be due to something else. To avoid this confounding factor in drug trials, one group of humans or pets is given a placebo— a substance or procedure a patient accepts as medicine or therapy, but which has no specific therapeutic activity. Any therapeutic effect from the placebo is due to the power of suggestion to the human or owner of the pet.
How would the power of suggestion affect a pet? If Fido’s human treats him differently because he thinks Fido is going to get better, that may be enough to help Fido improve. For instance if the dog suffers from fear of sounds and noises on walks and starts on medications the owner may feel more confident and stop acting nervous himself. The owner’s newfound confidence could then cause Fido to feel more relaxed.
Back to the Original Example
Now let’s go back to the original questions of training method A vs. B. Well the study clearly could not be double blind, because no matter what, the dog can tell which treatment he’s getting. And even if each dog had a separate trainer, the trainer would know which treatment the dog was getting too.
Could the study be blind? In most cases, no, but it depends on how the training is actually done. For one, both sets of dogs would need to wear an electronic collar but the clicker training group would have to have no experience getting shocked and would never get shocked during the course of the study.
The problem is that in the clicker training situation the trainer might be going back frequently to reward the dog as it’s lying down (e.g., if the click means a treat is coming but continue performing the same behavior) and if the dog starts getting up, the response of the trainer in the clicker training method would be different from that of the trainer training dogs in the second method.
How to decrease bias?
So how would you decrease bias? Besides being sure that the observers are trained well to recognize the behaviors, you could make sure the observers did not know the purpose of the study, had no biases against either method—(possibly because they are unfamiliar with training methods), and you could have multiple observers, only using results if you had good reliability between observer results.
If it sounds tricky to you, that’s the nature of behavior studies and what makes it tricky compared to other sciences such as biochemistry or genetics. But it’s also part of what makes behavior interesting and fun.
To read a blog summarizing an experiment that involved looking at the effects of using shock collars in training see Are Electronic Shock Collars Painful or Just Annoying to Dogs? A New Study Reveals Some Answers.
To read about an interesting study done by Dr. Yin on how dog barks differ depending on what they are saying go to Barking in Dogs: Noise or Communication?