Mary, You said: >...even a false correlation, given large enough numbers, would be worthy of further study..... Let's agree to disagree. I believe a mechanism for a causal connection between the events is needed to justify further study. And the mechanism must be plausible. Without a mechanism the numerical correlation-no matter how large-is of limited value. A large numerical correlation only suggests where to look for a mechanism. But, you knew you wanted to find a connection between a frozen shoulder and PD before you started the survey. Baldwin, 63/4 8x25/100 carbi/levo-dopa, 5 mg Eldepryl per day [log in to unmask] ---------- From: Mary Sheehan Sent: Saturday, October 25, 1997 10:06 PM To: Multiple recipients of list PARKINSN Subject: Answer to Survey Objections In a message dated 97-10-24 13:16:07 EDT, you write: << I responded to your survey as soon as I read it because it took me only a few minutes. I did it, however, even though surveys of this kind are usually not reliable. It is not that the questions you asked were deficient in any way, and I believe you would be objective in evaluating the answers.>> Thank you for replying to the survey in spite of your misgivings. << It is because retrospective studies are usually not a reliable way to establish correlations. I say usually because if everyone on the list answered and gave the same answer, you would indeed have uncovered something of value. But, results seldom turn out that way. Usually the correlations one might observe in retrospective studies have no validity and are useful only as a suggestion of one thing to look for in a more reliable subsequent study. >> I think a retrospective survey would have had some value because what I was looking for were patterns based on experience. Obviously, I am not equipped to do a blind study or a study lasting several years. What I was hoping to do was find, as you stated, "a suggestion of one thing to look for in a more reliable future study. " Interestingly enough, when I was at the Parkinson Institute for an appointment two weeks ago, I was asked to fill out a retrospective survey on hallucinations, so I couldn't be the only person who thinks such studies have value. <<Feller said that if you consider 100 uncorrelated random variables generated by a computer and calculated the correlations between all pairs of them, you would find at least one pair with a convincingly large numerical correlation. >> I understand this, but even a false correlation, given large enough numbers, would be worthy of further study..... or the correlation might have proved true and we just might have discovered something important. Mary