Experiments evaluated using multivariate methods Separating the effect of (correlated) explanatory variables Variation partitioning A A in addition to B B A or B B in addition to A In fact, more often used in observational studies - in experiments, we try to avoid correlated predictor (however, in ecology, we are not able to control everything) Effect of nitrogen fertilization on weed community Dose of fertilizer Cover of barley Weed community Pysek P. & Leps J. (1991):Response of a weed community to nitrogen fertilizer: a multivariate analysis. J. Veget. Sci. 2: 237-244. Effect of nitrogen fertilization on weed community Dose of fertilizer Cover of barley Weed community Basic questions: * Is there an effect of fertilization on the structure of the weed community? (either direct, or mediated through cover of the crop) Problem of correlated predictors: * Is there a direct effect of fertilization (i.e., which could not be explained as mediated through the cover of the crop)? * Is there an effect of the crop that can not be explained by the direct effect of fertilizer? Multiple regression: test of the complete model & test of partial (conditional) effects [plus possible test of marginal (simple) effects] Analysis of Variance; DV: NSP (fertenv.sta) Regress. Residual Total Sums of Squares Df 382.9215 2 394.6195 119 777.541 Mean Squares F 191.4607 57.7362 3.31613 p-level 2.98E-18 Regression Summary for Dependent Variable: NSP (fertenv.sta) R= .70176746 R2= .49247757 Adjusted R2= .48394778 F(2,119)=57.736 p<.00000 Std.Error of estimate: 1.8210 St. Err. St. Err. BETA of BETA B of B t(119) p-level Intercpt 9.423662 0.388684 24.24506 0 DOSE -0.02342 0.099678 -0.08501 0.361781 -0.23498 0.814629 COVER -0.68390 0.099678 -0.06174 0.008999 -6.86113 3.28E-10 +0.9 dose Stellaria media cover Veronica persica Galium aparine Myosotis arvensis -0.5 Anagalis arvensis Arenaria seryllifolia Fallopia convolvulus Vicia angustifolia Veronica arvensis Medicago lupulina Thlaspi arvense -1.0 +0.7 Note: in this Figure, CaseR scores are used instead of CaseE dose cover -1.0 +1.0 Variation partitioning Dose Cover Dose in Cover Cover in addition or addition to Cover Dose Dose A adjusted Variation partitioning - n.b. • In linear methods, trace (all the eigenvalues together) sum up to 1, so the eigenvalue corresponds to the proportion of explained variability • In unimodal methods, trace is higher than one, so the eigenvalue has to be divided by trace to get the proportion of explained variability Variation partitioning - n.b. • The variation could be partitioned among more than 2 variables (however, for more than 3 the clarity of the result is lost) • More useful: partitioning between groups of variables • The amount of explained variability is positively dependent on the number of explanatory variables in a group Spacková I., Kotorová I. & Leps J. (1998): Sensitivity of seedling recruitment to moss, litter and dominant removal in an oligotrophic wet meadow. Folia Geobot. Phytotax. 33: 17-30. Effect of dominant species, moss and litter on seedling germination Randomized complete blocks Ohrazeni 1994-seedlings, design of experiment (i3,8(f3.0)) 8 1 1 0 0 0 1 0 0 0 2 0 1 0 0 1 0 0 0 3 0 0 1 0 1 0 0 0 4 0 0 0 1 1 0 0 0 5 1 0 0 0 0 1 0 0 6 0 1 0 0 0 1 0 0 7 0 0 1 0 0 1 0 0 8 0 0 0 1 0 1 0 0 9 1 0 0 0 0 0 1 0 10 0 1 0 0 0 0 1 0 11 0 0 1 0 0 0 1 0 12 0 0 0 1 0 0 1 0 13 1 0 0 0 0 0 0 1 14 0 1 0 0 0 0 0 1 15 0 0 1 0 0 0 0 1 16 0 0 0 1 0 0 0 1 control litter-rnardus-rlit+mossblock1 rel1 rel2 rel3 rel4 rel5 rel11 rel12 rel13 rel14 rel15 Just of historical interest (the FORTRAN format etc.) block2 bl. rel6 re. rel16 Figure Chyba! V dokumentu není ΕΎádný text v zadaném stylu.-1: Environmental data characterizing the design of the experiment (Canoco file in full format). Standardization by cases Case1 Case2 1 10 5 50 7 70 10 100 3 30 If “standardize by case norm” is used, the two cases are identical Grubb theory of regeneration niche: importance of standardization - the standardization fundamentally changes the ecological interpretation of results If there are very different eigenvalues of the two displayed axes, then the “Focus scaling on” really plays a role! Note: centroids are scaled as cases on interspecies correlation on intercase distances Hierarchical structure each whole-plot is subdivided into 25 split-plots Seedlings - nested design [seme96su.spe, seme96su.env] 1 6 11 16 21 2 7 12 17 22 3 8 13 18 23 4 9 14 19 24 5 10 15 20 25 26 31 36 41 46 27 32 37 42 47 28 33 38 43 48 29 34 39 44 49 30 35 40 45 50 Permutations of the whole-plots 1 6 11 16 21 2 7 12 17 22 3 8 13 18 23 4 9 14 19 24 5 10 15 20 25 26 31 36 41 46 27 32 37 42 47 28 33 38 43 48 29 34 39 44 49 30 35 40 45 50 Repeated observations from a factorial experiment fertilization, mowing, dominant removal] 3 replications, i.e. 24 plots together Ohrazení (http://mapy.atlas.cz) Molinia caerulea Nardus stricta Species diversity and “interesting plants” (e.g. red list species) concentrated in “traditional”, i.e. mown, unfertilized Dactylorhiza majalis Senecio rivularis 14 Carex species Carex pulicaris C. hartmanii Summary of all Effects; design: (ohrazenv.sta) 1-MOWING, 2-FERTIL, 3-REMOV, 4-TIME 1 2 3 4 12 13 23 14 24 34 123 124 134 234 1234 df Effect 1 1 1 3 1 1 1 3 3 3 1 3 3 3 3 MS Effect 65.01041 404.2604 114.8438 87.95486 0.260417 213.0104 75.26041 75.53819 174.2882 41.48264 6.510417 14.67708 11.48264 2.565972 3.538194 df Error 16 16 16 48 16 16 16 48 48 48 16 48 48 48 48 MS Error 40.83333 40.83333 40.83333 7.430555 40.83333 40.83333 40.83333 7.430555 7.430555 7.430555 40.83333 7.430555 7.430555 7.430555 7.430555 F 1.592092 9.900255 2.8125 11.83692 0.006378 5.216582 1.843112 10.16589 23.45561 5.58271 0.159439 1.975234 1.545327 0.345327 0.476168 p-level 0.225112 0.006241 0.112957 6.35E-06 0.937339 0.036372 0.193427 2.69E-05 1.72E-09 0.002286 0.694953 0.130239 0.214901 0.792657 0.700348 Time • In repeated measures – time is categorial (but, linear or polynomial trends – contrasts – can be used) • In CANOCO, we can decide and use time either as a categorial or as a quantitative variable • If quantitative – we expect a trend! Interaction – just multiplication of the two values Time 0 1 2 Control 0 0 0 0 Treatment 1 0 1 2 Baseline: time=0 Note: expl. variables (incl. 3 interactions) 0 are centered 3 and standardized, but only after Treatment calculation of interactions) Control Time Time as A.D. Time Control Treatment 2000 2001 2002 2003 0 0 0 0 0 1 2000 2001 2002 2003 Treatment Control Time Time vs. Time * Treatment • Time: 0, 1, 2, 3 and 2000, 2001, 2002, 2003 – after centering and standardization, both series are identical • Time * treatment interaction – the results are very very different! C2 Explanatory variables Yr, Yr*M, Yr*F, Yr*R Yr*M, Yr*F, Yr*R C3 Yr*F C4 Yr*M C5 Yr*R Analysis C1 Covariates % expl. 1st axis st r 1 axis F ratio P PlotID 16.0 0.862 5.38 0.002 Yr, PlotID 7.0 0.834 2.76 0.002 6.1 0.824 4.40 0.002 3.5 0.683 2.50 0.002 2.0 0.458 1.37 0.040 Yr, Yr*M, Yr*R, PlotID Yr, Yr*F, Yr*R, PlotID Yr, Yr*M, Yr*F, PlotID After „subtraction“ of the covariate effect Original data Plot time1 time2 time3 time4 mean time1 time2 time3 time4 1 5 3 2 2 3 2 0 -1 -1 2 17 12 10 8 11.75 5.25 0.25 -1.75 -3.75 3 22 26 20 15 20.75 1.25 5.25 -0.75 -5.75 4 6 4 0 0 2.5 3.5 1.5 -2.5 -2.5 4.0 Principal response curves aulapalu 3.0 0.8 MR 2.0 M 1.0 R PRC1 FMR rhitsqua nardstri anthodor prunvulg carepilu brizmed luzumult carepale hylosple festovin poteerec selicarv carepani climdend ranuacer succprat planlanc pseupuru scorhumi lychflos careumbr brachyte carepuli ranunemo siegdecu triangles - mown circles unmown FM 0.0 -1.0 FR F full symbol - fertil. open symbol - unfert festrubr poa triv solid line - control descespi -2.0 -0.6 1994 1996 1998 2000 2002 2004 2006 YEAR broken l. - removal Further use of ordination scores Do we need PIC here?