BEES2041 Lecture Notes - Lecture 3: Ig Nobel Prize, Fisheries Science, Generalized Linear Model
9 Jun 2018
School
Course
Professor
The correct design (for testing temperature effects )
-replicate tanks within each treatment
-tanks randomly arranged in the lab
Non-independence can arise via:
-measuring same individual/object twice
-analyses differ when observations are paired vs. independent
A two sample test with paired samples:
-a fisheries scientist wanted to compare swimming speeds of fish at 2 temperatures (20 C
and 25 C)
-are swimming speeds different between two temperatures
Paired design
The data are paired because 2 measurements come from same individual fish
An independent two sample design for same question
Each measurement comes from different individual
Paired t-tests ask questions about size of the differences, not raw data
Why use paired sample designs
1 to reduce variance among individual measurements
2 e.g. test effect of drug on pairs of people matched by age & sex
3 reducing variance results in a more powerful test (i.e. greater ability to detect differences)
4 need to choose whether to pair observations prior to data collection
Non-independence can arise via:
-measurements being linked in any way
2007 Ig Nobel prize for nutrition: shows dung beetles are fussy eaters
-a comparison with t-test not valid as amounts eaten are not independent: beetles cant eat
2 things at once
-problems with independence are issues of logic & experimental design
-cannot be fixed by fiddling with the statistics
-must be considered before collecting data
Assumption 2
Samples are from normally distributed populations
-t-tests & some other statistical tests assume normally distributed data
-our ability to measure probability relies on known distributions
How do we check normality
-easiest method: plot frequency histograms
Assumption 3
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
The correct design (for testing temperature effects ) Analyses differ when observations are paired vs. independent. A fisheries scientist wanted to compare swimming speeds of fish at 2 temperatures (20 c and 25 c) The data are paired because 2 measurements come from same individual fish. An independent two sample design for same question. Paired t-tests ask questions about size of the differences, not raw data. 2 e. g. test effect of drug on pairs of people matched by age & sex. 3 reducing variance results in a more powerful test (i. e. greater ability to detect differences) 4 need to choose whether to pair observations prior to data collection. 2007 ig nobel prize for nutrition: shows dung beetles are fussy eaters. A comparison with t-test not valid as amounts eaten are not independent: beetles cant eat. Problems with independence are issues of logic & experimental design. Cannot be fixed by fiddling with the statistics.
