Evaluating Complex Survey Designs: A Simulation Approach

1
Evaluating Complex Survey Designs A Simulation
Approach
S. Garman (NCPN) C. Lauver (SCPN) B. Schweiger
(ROMN) T. Mau-Crimmins (SODN) R. Bennetts
(GRYN) D. Manier (ROMN) E. Crowe (GRYN) With
assistance from S. Urquhart, P. Larsen, T. Kincaid
2
The Issue

• High degree of uncertainty in real-world
• variance of Vital Sign(s)
• But, may have estimates of variability from
• approaches similar to those proposed for
• monitoring a VS
• Use these estimates (or guesses) to explore
• the performance of alternative survey
• designs, given assumptions of required
• change-detection levels per unit time

3
A Simulation Approach

• A simulation approach offers the ability to
  evaluate performance among many sampling designs
  under various assumptions of plot-level
  variability
• CSDSim model
• Simulates trends on plots distributed
• across a landscape (the sampling frame).
• Variability estimates used to control initial
• conditions, slope, and RootMSE of obs.
• around the slope on a plot-level basis.

4
Simulate the population
For each plot, generate initial indicator
value, slope over specified time, annual
observations around the slope. User-specified m
u SD initial conditions mu SD slope mu
SD RootMSE Derivation of values value mu
SD n.r.v
5
Simulated observations for 2 plotsNotice
different initial values, slopes, RootMSE
6
Apply CSDs to the landscape E.g.,

(2-7)9, (1-8)9 X 3, 5, 7 .. plots
(1-0)1, (2-2)4, (1-3)4 X 3, 5, 7 .. plots
7
Extracted Samples (e.g., 2-2 design)
8
Apply CSDs to the landscape

• Plot locations are either randomly determined
• or user provided (e.g., GRTS sample plots
• referenced by row column at least k plots
  must be
• provided where k is defined by the sampling
  design)
• Monte Carlo approach used to acquire samples
  using
• different combinations of plots
• if random, new plots selected each replication
• if user-provided, must have n number of plot
  sets. E.g.,for
• n versions of GRTS samples, each version is
  generated using a
• different random number seed.

9
Analyses

• Derive variance components of sample
• observations for Power-for-trend assessments
• Use Mixed-effects
• model to generate
• site,year,
• siteyear,
• index variances
• Power.fcn.R derives
• power to detect
• specified trend
• with results
• from mixed-
• effects model
• re-visit design
• (T. Kincaid)

10
Analyses

• Compare sample slope and status with population
  slope and status determine proportion of 100
  replicates not sign. different from population
  parameters

• trend - Mixed-effects model (type 3)
• time fixed effect (continuous)
• site, year random effects (cat.)
• slope estimate, sign. of slope, variance
• of slope, Residual MSE

11
Analyses

• status B0 B1 time
• where time the most recent time
• - variance is from the cov matrix CVC,
• where,
• V cov matrix of regression coeff.
• C is a 2x2 matrix
• 1 time
• 0 1

12
Population slope -0.2 of sample replicates
with a slope estimate not-sign. different from
population slope at yr 10
of sample replicates providing a status
estimate not-sign. different from population
status at yr 10
Alpha 0.05
3,7,11,15
13
3,7,11,15 4,5,6,8,9,10 3,5,7
14
Enhancements under development
User specified spatial pattern of initial
conditions, slope, RootMSE, year effects
Off-line creation of spatial pattern of initial
values. Initial values read into memory at
program initiation.