Request of feedback on planned proceedure

[pablo arnal]

Dear members,

I am writing you to request your feedback on the procedure I am planning to perform to estimate sample size and to ask you a few questions.

I am researching how to carbonize biomass -like sunflower hulls- by simple methods at laboratory-scale. Recently, I found a new simple method (let us call it method A). On the other side, in the scientific literature we can find that scientist have used so far other 3 different methods to carbonize biomass at laboratory-scale (let us call them methods B, C & D).

When trying to find out which percentage of published articles used methods B, C & D, respectively, I faced a big problem: 5300 articles to check. I ought to check each article (found with a set of keywords in Scopus) to determine whether method B, C or D has been used has been used in each article.

Instead of reading all of them, I would like to determine the number of articles that I should randomly select (i.e. the sample size) to be able to determine the percentage of articles that use methods B, C and D in the literature with an accuracy of 10%.

To obtain this estimate for each method, I used Russ Lenth's applet (CI for a proportion) with following input:

• finite population: 5300
• worst case: 0.8
• confidence: 0.9
• Margin of error: 0.1

As result, I obtained a sample size of 44 (43.94). 

A few questions:

• Why does the check mark dissapear when I change the worst case from .5 to .8?
• Does worst case mean power of test (i.e. pi)?
• Does margin of error mean accuracy?
• Does confidence 0.9 means alfa=0.1?
• Does 10% accuracy mean that the experimentally obtained value for the proportion has an error of +/-5%?
• Should I use different sets of 44-randomly generated numbers (from 1 to 5300) for each type of method of carbon preparation?

Your feedback will be highly appreciated.

Kind regards

Pablo

[Lenth, Russell V]

The worst case is when the true pi is .5. That's the case where the CI has maximum width. The check mark goes away when you pick a different value of pi. The 10% margin of error means your confidence interval is the observed percent, plus or minus 10 percent points. You will need about 4 times as much data to get plus or minus 5 percent. In addition, I suggest you use 95% confidence, i.e. alpha = .05. You should use only one sample, and tally the percent of each method found in the studies you sampled.

Russ

Sent from my iPad

--
You received this message because you are subscribed to the Google Groups "PiFace discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to piface-discuss...@googlegroups.com.
To post to this group, send email to piface-d...@googlegroups.com.
Visit this group at http://groups.google.com/group/piface-discussion.
To view this discussion on the web visit https://groups.google.com/d/msgid/piface-discussion/26e58a01-3c5c-4584-9a15-f52304980d9a%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.