Ultimately, science is about cause and effect. Psychological scientists want to be able to understand and predict behaviors and mental processes.从 ::27 开始播放视频并学习脚本0:27In seeking that understanding, researchers are concerned with a form of validity called internal validity. A study has high internal validity when observed changes in the outcome, known as the dependent variable, are likely due only to changes in a manipulated variable, known as the independent variable.从 ::48 开始播放视频并学习脚本0:48This allows researchers to draw reasonable conclusions about a cause and effect relationship between the independent and dependent variables. Achieving this requires that there be no confounding variables that affect the dependent variable, that the independent variable be entirely under the researcher's control, and that it is possible to entirely control extraneous variables.从 :1:14 开始播放视频并学习脚本1:14The only research design that can achieve high internal validity is the experimental design. In an experimental setting, researchers are able to control extraneous variables and manipulate an independent variable to determine if and how changes in the independent variable lead to changes in a dependent variable.从 :1:36 开始播放视频并学习脚本1:36In contrast, in a correlational design, researchers do not have this kind of control over extraneous and independent variables, so they can only observe associations between variables.从 :1:50 开始播放视频并学习脚本1:50However, not even all experiments have high internal validity. Studies are seldom perfect, nor are researchers, so problems sometimes creep in, even in true experiments.从 :2:4 开始播放视频并学习脚本2:04For example, in nineteen-forty-nine the journal Psychological Review published a paper by Elliott McGinnies in which McGinnies claimed to demonstrate a phenomenon he called perceptual defense. His idea was that the subconscious mind will delay conscious perception of a stimulus that it finds threatening.从 :2:24 开始播放视频并学习脚本2:24McGinnies set up an experiment that displayed a word for one-hundredth of a second, after which the participant would be asked to name the word, if possible. If not, the duration would be increased by another one-hundredth of a second, the word would be presented again, and the participant would be given the opportunity to name it if possible. This procedure was repeated until the participant correctly named the word. The dependent variable was the exposure duration at which the participant correctly named the word.从 :2:59 开始播放视频并学习脚本2:59The independent variable was the type of word displayed. Some words were emotionally neutral, like apple or dance. Other words were emotionally charged words that were socially taboo in nineteen-forty-nine, like belly or bitch. McGinnies called this set the critical words.从 :3:19 开始播放视频并学习脚本3:19On average, the neutral words took just over five hundredths of a second to recognize, but the critical words took nearly ten hundredths of a second to recognize. McGinnies interpreted this difference as evidence that unconscious parts of the mind were defending against conscious perception of the critical words.从 :3:40 开始播放视频并学习脚本3:40However, there is another interpretation. Davis Howes and Richard Solomon suggested in a nineteen-fifty article in Psychological Review that the difference in the frequency with which people see and hear the various words in the neutral and taboo sets might account for the effect observed by McGinnies. They cited research showing that the frequency with which a word is used in newspapers, magazines, and books is related to the threshold amount of time needed for recognition. As shown in this graph, the more commonly a word occurs in print, the less time it takes to recognize and correctly name the word.从 :4:22 开始播放视频并学习脚本4:22Using those results, Howes and Solomon then noted that the words in McGinnies' critical set were all less frequent than the words in the neutral set, so based on word frequency alone, we would expect longer times for recognition of the critical words than for the neutral words, as shown in this graph where the estimated recognition times for the specific words McGinnies used, are plotted against those words' frequency in the language. This creates a situation where there are at least two variables that can plausibly explain the change in the dependent variable, either the type of meaning of the word, emotional or neutral, or the frequency of occurrence of the word.从 :5:5 开始播放视频并学习脚本5:05This kind of problem is known as confounding. In this study, word frequency is a confounding variable. In general, an experiment has a confounding variable when an uncontrolled variable is correlated somehow with the independent variable, so that the confounding variable adds another possible explanation for any differences observed in the dependent variable.从 :5:30 开始播放视频并学习脚本5:30In the case of McGinnies's experiment, the threshold time for recognition was higher in the critical condition than in the neutral condition. It might be that the content of the words affected recognition time, with taboo content causing an increase in recognition time, relative to neutral content. Alternatively, it might be that the natural frequency of experience with the words affected recognition time, with words encountered less frequently causing an increase in recognition time, relative to more frequently occurring words. And of course, the difference in the dependent variable might be due to some combination of these factors. There is simply no way to know when there is a confounding variable.从 :6:16 开始播放视频并学习脚本6:16Confounding variables are sometimes subtle and experts might not identify them for some time. For example, not too long ago, medical experts concluded that a moderate level of alcohol consumption is good for your health. Many studies had reported results in which the risk of contracting a life-threatening disease–particularly heart disease–appeared to be lower for people who consumed moderate amounts of alcohol, relative to people who consumed little or none. However, more recent interpretations suggest that the early studies were confounded because the nondrinkers were more sick on average to begin with, which is why many of them were nondrinkers in the first place.从 :7:1 开始播放视频并学习脚本7:01The more recent interpretation suggests that when we account for this confounding variable, the data showed that any alcohol consumption increases the risk of heart disease. For relatively low levels of alcohol consumption, the increase is smaller than for higher levels of consumption. The important point here is that there is an increase in risk, not a decrease as was once believed. When thinking about cause and effect relationships between variables, we have to carefully consider other variables that might confound that relationship.从 :7:37 开始播放视频并学习脚本7:37This module introduces the ideas of internal validity and confounding and then covers various steps to avoid or minimize problems from confounding variables in experimental designs.