Upper Mississippi River Restoration Program Long Term Resource Monitoring

Estimating means and temporal trends using LTRM data: details with examples

Estimating temporal trends in data from multiple strata

Estimating temporal trends from data from multiple strata requires adjustment for variation in sampling probabilities.  As discussed earlier, users may also choose to adjust for stratification effects within years. We illustrate the estimation of temporal trends with and without adjustment for stratification effects below. 

Estimating temporal trends in fish data from multiple strata

As we did with the single strata trend model of count data, we first preprocess data in SAS prior to submitting those data to SUDAAN. In this case, we add a centered year variable, define a large number (10,000) of possible years from which to sample, create an effort variable in units of 15 minutes, and exclude data from period 1.

data BLGLP8wtb;
set BLGLP8wt;
yearctr = year - 2000;
yearcnt = 10000;
effmin15 = effmin / 15;
where period gt 1 and fstation = 2;
run;
proc sort data = BLGLP8wtb; by year stratum_code; run;

We then tell SUDAAN that the input file will be of type "sas," and that the sampling design is without replacement ("wor").  This method allows trend estimation with correction for correlation within years (‘nest’ statement) and sampling weights (‘weight’ statement). The nest statement defines a single stratum in which years occur ("_one_"; in this case, corresponding to Pool 8), that years represent primary sampling units and that LTRMP’s spatial strata are being treated as secondary strata. The "totcnt" statement defines the number of potential sampling units from which years were presumably drawn ("yearcnt"), a stratification variable (_zero_) and that we used with replacement sampling for observations within the secondary strata ("_minus1_"). The model and print statements tell SUDAAN what we are modeling and how we want results formatted.

Temporal trends in fish detection/nondetection data may be estimated using code supplied for estimating percent frequency of occurrence in the vegetation example below.

Estimating temporal trends in macroinvertebrate data from multiple strata

Temporal trends in macroinvertebrate counts from multiple strata may be estimated using the code supplied for bluegills above. Trends in macroinvertebrate detection/nondetection data may be estimated using code supplied for estimating percent frequency of occurrence in the vegetation example below.

Estimating temporal trends in vegetation data from multiple strata

Temporal trends in vegetation data from multiple strata may be estimated by modifying the vegetation code for single strata. As there, we use wild celery data from the impounded stratum of Pool 13 and begin by creating a centered year variable.

data VAAM3FS3wtb;
set VAAM3FS3wt;
yearctr=year-2000;

Trends in percent frequency of occurrence may be estimated using the following code

proc surveylogistic data=VAAM3FS3wtb;
cluster year;
model sitedetect = yearctr / s;
weight sweight;
where pool = “13”;
run;

The above code may be modified by adding a class statement (e.g., "class cov_nrf;") to estimate a common trend in categorical cover classes.

The code in this vegetation section does not adjust for stratification effects. Users who wish to adjust for the effects of stratification on temporal trend estimates may adapt the LOGLINK code supplied above for bluegills for use with SUDAAN's LOGIT procedure.

Estimating temporal trends in water quality data from multiple strata

Temporal trends in water quality data from multiple strata may be estimated using the SURVEYREG procedure used above for trends in vegetation data. Here we illustrate estimating temporal trends in chlorophyll a in Pool 13. As typical we center year by subtracting a value within the sampling period.

data WQallwt2; set WQallwt; yearctr=year-2000;

proc surveyreg data=WQallwt2;
cluster year;
model chlf = yearctr / s;
weight sweightstd;
where fs=3;
run;