If you are performing an experiment in which you tell the participant what you are expecting, then this biases the result due to placebo effect. This is the observer expectancy effect. How can you measure this? Intuitively, the solution seems simple. Do the experiment twice, in the first, tell the participant about the expected outcomes and in the second don't tell him anything about the experiment. Then, you measure the difference in outcomes to calculate the variance introduced due to observer expectancy effect. But, is that accurate?
Lets consider a simple example. Suppose you are to create stress relief program. How would you measure if stress is relived or not? If you tell your subjects that they were participating in stress relief program, placebo effect will come into account and you cannot truly determine if the reduction in stress is actually due to the program you created. If you don't tell anyone about anything, including researchers and participants, you can get rid of the observer expectancy effect. This is the Double Blind trial strategy.
Coming back to the original question, if you do the experiment twice, one with the expectancy effect and another using double blind strategy and consider the difference in performance, do we then have the measure of observer expectancy?
The answer is NO because in both the experiments the state of the participant is different. To be accurate, you'll have to conduct both the experiments in which the researchers, participants and in fact the entire universe is in the same state, i.e., do both the experiments simultaneously, which obviously doesn't work out.
How else can we go about this problem? First we start by formalizing the problem, making it concise. For simplicity, let us consider a single participant. In a given experiment, let the state of the participant be Sp (could involve factors such as personality etc..) and the state of everything else be Se (state of the environment, ideally the entire universe, but a local region would suffice). Therefore, the state of an experiment can be defined by the Tuple (Sp, Se).
Now, perform N experiments with blind trial strategy, each represented by different tuples (S1p, S1e) ... (Snp, Sne) You can now build a regression model to determine the the outcome of the experiment as a function of Se and Sp, after collecting data from sufficiently large number of experiments.
Now we can apply the strategy discussed before. We perform experiment with double blind trial with parameters (S1e, S1p). The second experiment (with expectancy effect) with parameters (S2e, S2p). We can now extrapolate the outcomes of first experiment if parameters S2e and S2p were used instead of S1e and S1p. Since, both the experiments are now virtually conducted simultaneously, we can now determine the observer expectancy effect by computing the difference in outcome.
More accurate the regression model, better is the accuracy of the observer expectancy. With few obvious modifications, one can also build a model to estimate observer expectancy as a function of experiment, Se and Sp.
On a second thought, who gives a damn? If the stress relief program works, be it due to expectancy, that's all we really care about.
Lets consider a simple example. Suppose you are to create stress relief program. How would you measure if stress is relived or not? If you tell your subjects that they were participating in stress relief program, placebo effect will come into account and you cannot truly determine if the reduction in stress is actually due to the program you created. If you don't tell anyone about anything, including researchers and participants, you can get rid of the observer expectancy effect. This is the Double Blind trial strategy.
Coming back to the original question, if you do the experiment twice, one with the expectancy effect and another using double blind strategy and consider the difference in performance, do we then have the measure of observer expectancy?
The answer is NO because in both the experiments the state of the participant is different. To be accurate, you'll have to conduct both the experiments in which the researchers, participants and in fact the entire universe is in the same state, i.e., do both the experiments simultaneously, which obviously doesn't work out.
How else can we go about this problem? First we start by formalizing the problem, making it concise. For simplicity, let us consider a single participant. In a given experiment, let the state of the participant be Sp (could involve factors such as personality etc..) and the state of everything else be Se (state of the environment, ideally the entire universe, but a local region would suffice). Therefore, the state of an experiment can be defined by the Tuple (Sp, Se).
Now, perform N experiments with blind trial strategy, each represented by different tuples (S1p, S1e) ... (Snp, Sne) You can now build a regression model to determine the the outcome of the experiment as a function of Se and Sp, after collecting data from sufficiently large number of experiments.
Now we can apply the strategy discussed before. We perform experiment with double blind trial with parameters (S1e, S1p). The second experiment (with expectancy effect) with parameters (S2e, S2p). We can now extrapolate the outcomes of first experiment if parameters S2e and S2p were used instead of S1e and S1p. Since, both the experiments are now virtually conducted simultaneously, we can now determine the observer expectancy effect by computing the difference in outcome.
More accurate the regression model, better is the accuracy of the observer expectancy. With few obvious modifications, one can also build a model to estimate observer expectancy as a function of experiment, Se and Sp.
On a second thought, who gives a damn? If the stress relief program works, be it due to expectancy, that's all we really care about.
No comments:
Post a Comment