On 3 Mar 2004 07:24:00 -0800, [EMAIL PROTECTED] (Euh) wrote: > "Jim Snow" <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>... > > "Euh" <[EMAIL PROTECTED]> wrote in message > > news:[EMAIL PROTECTED] > > > Hello all, > > > > > > I'm trying to evaluate the concentration of suspended particles �n a > > > flask. > > > I took 3 differents samples from the flask and, for each sample, I did > > > 20 counts under the microscope (= at 20 different locations on the > > > microscope slide) > > > > (snip) > > > > > > > Data (each column is a sample, each row is a count) > > > > > > 98 108 115 > > > 123 102 120 > > > 88 90 93 > > > 65 71 92 > > > 95 56 131 > > > 68 145 138 > > > 114 136 116 > > > 82 100 98 > > > 87 85 70 > > > 109 116 56 > > > 134 157 34 > > > 102 60 113 > > > 130 90 125 > > > 53 121 63 > > > 124 111 81 > > > 114 92 117 > > > 118 131 109 > > > 76 55 113 > > > 97 256 108 > > > 79 146 72 > > > > Your counts ,I think, are supposed to be counts of the number of > > particles in equal volumes of a well mixed suspension. If this were > > true, the numbers would follow a Poisson distribution, characterised > > by the population variance being equal to the mean. This is clearly > > not true for your samples: > > > > Means are 97.8, 111.4, 98.2 Variances are 527, 2061, 760 > > > > Either the volumes contributing to each count are variable or mixing > > has not been thorough enough, or I do not understand your description > > of the data. > > > > HTH Jim Snow > > It's not a problem of mixing, but it's related. Particles tend to form > aggregates which makes it harder to count the total number. > On top of that, some particles are degraded over time and you end up > with debris in the solution. Sometimes, it's hard to tell if what you > see is a small particle (that should be counted) or a debris (should > be ignored). it's judgment based...hence the huge variation. > > By adding one count at a time and computing a t-value, I've been able > to evaluate that approx 10 counts are required to get close to the > true value (15 % error).
I find myself wondering about "stratified sampling" or "blind (or mechanical) selection" -- because those 20 values offered are highly inconsistent. It is not possible that they represent a single, homogeneous mean, but they *could* represent a single "totality" if they do represent strata, in the style of stratified-sampling. One curiosity in the numbers as they have been described, as I see it, is that the 3 samples ended up with averages so similar. I think it must be an accident, unless it represents an unconscious *bias* to represent different amalgams with equal fractions. - I conclude that I want to see an explanation of the sampling, before I believe that the results are not confounded or biased by 'rater' effects that could be sizable. > > > I've also computed the number of samples required based on the > standard deviation of all the counts and this gives me a higher > number. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html - I need a new job, after March 31. Openings? - . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
