Strategies for the Collection and Use of Quantitative Data

1
Strategies for the Collection and Use of
Quantitative Data
Norma Fowler Section of Integrative
Biology University of Texas at Austin
2
Why are quantitative data not used as widely and
effectively as they might be?limited
funding......so...How can we obtain and use
quantitative data most efficiently, given
inevitable funding constraints?
3

• Some reasons why quantitative data are not used,
  or are used incorrectly or inefficiently (other
  than funding limitations)
• data not collected
• data collected but not analyzed
• inefficient sampling and experimental designs
• statistically invalid sampling and experimental
  designs

4

• Some reasons why quantitative data are not used,
  or are used incorrectly or inefficiently (other
  than funding limitations)
• data not collected
• data collected but not analyzed
• inefficient sampling and experimental designs
• statistically invalid sampling and experimental
  designs

5

• data not collected the lack of long-term
  monitoring
• long-term monitoring
• reveals whether a management practice is working
• provides baseline data for future studies and
  future management
• is often relatively inexpensive
• but....
• it is not as glamorous as new initiatives
• the need often outlasts the duration of project
  funding, the employment of a staff member, etc.

6

• Some reasons why quantitative data are not used,
  or are used incorrectly or inefficiently (other
  than funding limitations)
• data not collected
• data collected but not analyzed
• inefficient sampling and experimental designs
• statistically invalid sampling and experimental
  designs

7
data not analyzed

• because bookkeeping and number-crunching arent
  as much fun as field work?
• because the funding has run out?

image from http//www.sostitle.com/
8

• Some reasons why quantitative data are not used,
  or are used incorrectly or inefficiently (other
  than funding limitations)
• data not collected
• data collected but not analyzed
• inefficient sampling and experimental designs
• statistically invalid sampling and experimental
  designs

9

• the three most common types of important design
  problems seem to be
• replication
• no replication
• no statistically valid replication
• inefficient replication
• randomization
• no randomization
• statistically invalid randomization
• number of statistical units v. number of variables

10

• the three most common types of important design
  problems seem to be
• replication
• no replication
• no statistically valid replication
• inefficient replication
• randomization
• no randomization
• statistically invalid randomization
• number of statistical units v. number of variables

11

• replication
• An example of invalid, valid but inefficient, and
  more efficient designs
• 3 habitat types
• open cluster along
• savanna of woody plants a drainage
  

for simplicity, assume we can only make
measurements on 36 plants
12
replication
valid but inefficient design 2 plots
per habitat type, 6 plants per plot
statistically invalid design only 1 plot per
habitat type
13
replication
alternate approach no plots, 12 plants per
habitat type stratified random plant
locations
statistically more efficient design 4 plots
per habitat type, 3 plants per plot
14

• the three most common types of important design
  problems seem to be
• replication
• no replication
• no statistically valid replication
• inefficient replication
• randomization
• no randomization
• statistically invalid randomization
• number of statistical units v. number of variables

15
randomization

• some really bad ideas
• ignore the whole issue and hope it goes away
• throw hoops with your eyes shut
• sample whenever there is a place to pull off the
  highway
• sample every 10 m
• wander around and sample at what feels like
  random intervals to you

16
randomization

• Humans dont seem to be able to generate random
  numbers.
• Solutions
• random number tables correctly used
• random number generators in computer packages

17

• the three most common types of important design
  problems seem to be
• replication
• no replication
• no statistically valid replication
• inefficient replication
• randomization
• no randomization
• statistically invalid randomization
• number of statistical units v. number of variables

18
number of statistical units v. number of variables

• Too many characters on too few statistical units
  ( plots or plants, depending on the design).
  Why?
• Big plants are big all over usually little is
  gained by measuring gt1 size-related variable.
• Environmental variables are often so highly
  correlated with each other that little is gained
  by measuring some of them.
• If plot is the statistical unit of replication,
  more plants per plot usually add little to the
  power of the analysis.

19

• An example of a statistically valid, relatively
  efficient design the effects of deer browsing on
  Streptanthus bracteatus

2 treatments fenced exclosure control
1 cluster of plants 1 plot
6 plots per treatment in yr 1, 9 plots per
treatment in yr 2
images from http//www.wildflower2.org/NPIN/Galler
y/ and http//www.thenorthview.org/recreation/anim
als/deer.html
20

• preliminary study of 11 size-related measures of
  Streptanthus bracteatus
• stem basal diameter predicted dry biomass very
  well (R2 0.95),
• the other 10 variables did not have
  to be measured in the experiment,
• and....
• using this estimate of initial biomass greatly
  increased the power of the analysis

image from http//www.tpwd.state.tx.us/news/tv/vnr
/archive/0307
21
(No Transcript)