Quantile Regression: EPT Relative Abundance vs. Percent Sands and Fines for Minnesota Streams

Plot

Analysis: Quantile regression
Independent variable: Percentage (%), sand and fines
Dependent variable: EPT (Ephemeroptera, Plecoptera and Trichoptera) abundance/m²
Classification variables: Ecoregion class - dummy variable = 0 if the Ecoregions are 46, 47, or 51 and dummy variable = 1 if the Ecoregions are 50 or 52
Key: symbols = observations, solid line = mean line, dashed line = 95% prediction limit, black lines and symbols = Ecoregions 46, 47 and 51, blue lines and symbols = Ecoregions 50 and 52
Sample size: n = 267

Variable	Estimate	Chi²	p
Intercept	341.6
Ecoregion class	-93.0	2.74	0.098
% Sand & fines	-1.88	6.08	0.014^a
% Sand & fines x Ecoregion class	1.75	2.66	0.10

^aHowever, Chi²had a p > 0.05 (Chi² = 1.09, p = 0.30) when "ecoregion class" was removed from the quantile regression.

Model generated using SAS Proc QUANTREG with the smoothing algorithm. The Wald test statistic was used to test the linear null hypothesis, H₀: slope=0, for all independent variables (i.e., ecoregion class, % sand & fines, and ecoregion class x % sand & fines.)
The ecoregions were aggregated into two classes because a preliminary ANOVA indicated that the base material for bedded sediments, % coarse substrates, was significantly greater in Ecoregions 50 and 52 than in Ecoregions 46, 47, and 51. Percent coarse substrates, however, did not significantly differ within either group.
Replicate visits to a site were excluded from the data set.

Agency: Minnesota Pollution Control Agency, St. Paul, Minnesota
Location: Minnesota - analyses restricted to wadeable sites (52 km² < watershed area < 523 km²) from the five ecoregions indicated below.
Omernik Level III Ecoregions examined include: 46 (Northern Glaciated Plains), 47 (Western Corn Belt Plains), 50 (Northern Lakes and Forests), 51 (North Central Hardwoods), and 52 (Driftless Area).

Michael B. Griffith, griffith.michael@epa.gov, (513) 569-7034