In language, as with any other complex skill, performance does not always relect internal knowledge, or competence. Just as a professional basketball player who knows very well how to shoot the ball from the free-throw line will occasionally miss the shot, perfectly competent speakers of English routinely fumble with their words or suddenly find themselves unable to retrieve a word they've used thousands of times. This sort of flub is purely a performance error and says nothing about the speaker's competence. The difficulty inherent in linguistic research is that all we can directly measure is performance. A given speaker's competence is always buried in the "black box" of the human mind and can only be estimated through the combined efforts of countless researchers and theoreticians.
For an example of how sharply linguistic performance can differ from competence, try to imagine your first date. Or better yet, try to imagine mustering up the courage to ask your high school crush to the prom. Of course you know how to talk perfectly well, but the words just don't come out right. Your heart is racing, your hands are sweating, and your throat is bone dry. All of these factors conspire to make you . . . less than eloquent. Or imagine you're in a meeting with your boss trying to figure out the best way to ask for a raise. Or giving a presentation to an auditorium full of your colleagues and superiors. Whenever your audience has the power to affect your life for better or worse, your anxiety level may rise and your performance decline.
This same phenomenon occurs (though perhaps to a much lesser extent) with many participants in linguistic experiments. In an experimental setting, when confronted by a credentialed researcher hanging on their every word (and hoping against hope for some "interesting" data), it's perfectly reasonable that some people will freeze up a bit. Apart from being overly concerned with grammatical correctness, participants may also attempt to give the researcher what they want to hear, making guesses about what the focus of the research is all the while. This makes linguistic research particularly tricky, even among the social sciences. But after all, if it were easy to gain a coherent picture of a given speaker's linguistic competence, research would not be necessary, and many of us would be out of a job.