Impact Assessment

1
Impact Assessment

• ESM 206C
• 3 April 2007

2
Impact assessment

• What is the impact of one or more management
  techniques on an environmental variable of
  interest?
• Effects of grazing on biodiversity in California
  grasslands
• Establishment of marine protected areas in
  Channel Islands how do they affect lobster
  abundance?
• Has some event caused a change in the
  environment?
• Power plant goes on line
• Hurricane
• Challenge is to establish causation
• This is a conceptual issue, not a statistical
  issue!

3
San Onofre Nuclear Power Plant

• Units 2 3 came online in May 1983
• Cooling water discharge creates turbidity due to
  volume of flow
• This has potential negative effects on giant
  kelp
• Reduced light at sea floor reduces growth and
  survival of baby kelp
• Sediments bury hard substrate
• Data collected on total area of kelp forest,
  using sidescan sonar
• Question did the power plants negatively impact
  the kelp bed adjacent to the cooling water
  discharge site?

4
Before-After (BA)

• Suppose we have data from power plant outflow
  site before and after operation began
• Is kelp area lower after the power plant went
  online?

5
Two-sample t-test

• If two samples, X and Y, are from populations
  with the same mean
• Then the quantity
• follows a t distribution with
  nx ny 2 degrees of freedom

6
Assumptions of 2-sample t-test

• Populations from which x and y are sampled are
  normally distributed
• Test is pretty robust as long as sample sizes are
  similar and 2-tailed tests are being considered
• The more samples, the better
• If non-normality is strong, dont put confidence
  in alpha levels below 0.01
• Populations from which x and y are sampled have
  same variances
• Violations mean that P values will be somewhat
  too small
• Use Welchs approximate t test instead

7
Welchs approximate t-test

• If two samples, X and Y, are from populations
  with the same mean but different variances
• Then the quantity
• follows a t distribution with
• degrees of freedom

8
BA (continued)
9
BA (conclusion)

• The assumptions of the t-test are moderately
  violated
• However, P value is extremely small we can be
  confident that the true P is less than a
  reasonable alpha (0.05 or even 0.01)
• Thus, with high confidence we can reject the null
  hypothesis that the kelp area was the same before
  and after the power plant went online
• Does this mean the power plant has caused this
  difference?

10
Control-Impact (CI)

• Suppose we have data from power plant outflow
  site and a control site nearby, but only after
  operation began
• Is kelp area lower in impact site than control
  site?

11
Again, do a 2-sample t-test
12
CI

• Conclusion Fail to reject the null hypothesis
  that Control and Impact sites have same amount of
  kelp!
• Does this mean that power plant has no effect on
  kelp?

13
Before-After Control-Impact (BACI)

• If we have data from both sites (Control and
  Impact) at both periods (Before and After) then
  we can
• Use the Control site to control for temporal
  changes in kelp that are unrelated to the power
  plant coming on line
• Also called a counterfactual
• Use the Before period to determine the relative
  quality of the two sites before the power plant
• We want to focus on the difference of
  differences

14
Estimate the effect size
15
Combining all the information simple BACI using
ANOVA

• Use Site, Period and Site x Period interaction in
  ANOVA
• The model

S site (C1, I-1) P period (B-1, A1)
Overall mean kelp density
Deviation from overall mean associated with being
in a given Period (After b2, Before -b2),
averaged over both sites
Deviation from site mean associated with being in
a given Period (After b3, Before -b3),
OR Deviation from period mean associated with
being in a given Site (Control b3, Impact -b3)
Deviation from overall mean associated with being
in a given site (Control b1, Impact -b1),
averaged over both periods
16

• Overall mean
• Site means
• Period means
• Interactions

17
BACI with ANOVA
18
We can do even better!
19
Before-After Control-Impact Paired Series (BACIPS)

• Calculate deltas for each sample time by
  subtracting value at control site from value at
  impact site
• controls for temporal variability in environment
• Do a two-sample t-test to see if mean delta
  changes from before to after

20
BACIPS continued
21
BACI BACIPS conclusions

• Reject null hypothesis that true effect size is
  zero
• BACIPS gives more power
• Could have drawn conclusion sooner
• Gives more confidence under violation of
  assumptions
• Does this mean the power plant has caused this
  effect?

22
Impact assessment general considerations

• We need some way to control for variability that
  might confound our conclusions
• Temporal changes unrelated to event in question
• Differences in underlying quality between sites
• Can also control by including measurements of
  additional variables that might affect response
• SST, amount of rocky reef, etc.

• What if monitoring only starts after event occurs?

23
Environmental challenge

• The problem There is a great deal of controversy
  about how to manage California grasslands for
  biodiversity values. In particular, there is
  debate about whether grazing by cattle is
  promotes or is deleterious to plant diversity
• Data on plant diversity and relative abundance
  have been collected from plots at Sedgwick
  Reserve where grazing has been either allowed or
  excluded.
• Your job Using these data, determine whether
  grazing increases or decreases plant diversity.

24
Diversity in Sedgwick grassland plots
25
2-sample t-test of H
26
Paired t-test

• If observations naturally come in pairs
• Control treatment plots next to each other
• Calculate differences, di xi - yi
• Use one-sample t-test to test H0 md 0

27
Paired t-test
28
2-sample vs paired t-test

• TWO SAMPLE t-TEST
• Difference of means
• Use when observations are independent between
  groups
• Assumes each population is normally distributed

• PAIRED t-TEST
• Mean of differences
• Use when observations are naturally paired
• Assumes population of differences is normally
  distributed

• Both will estimate same mean difference
• If there is variation among pairs (e.g., due to
  location, soil type, habitat), then paired test
  will have more power to reject null hypothesis

29
Independence and randomness
30
Further Reading

• Schmitt, R.J., and C.W. Osenberg, eds. 1996.
  Detecting Ecological Impacts. Academic Press, San
  Diego.
• Stewart-Oaten, A., W.W. Murdoch, and K.R. Parker.
  1986. Environmental impact assessment
  pseudoreplication in time? Ecology 67 929-940.
• Zar, Chapters 8.1-8.4, 9.1-9.3 (t-tests), 10.l,
  12.1-12.2 (ANOVA)