C. Stoate et al. Agriculture, Ecosystems and Environment 77 2000 219–226 223
considered exclusively. Variables entering the mod- els winter and spring were selected through forward
stepwise selection. In the second stage, 114 transects and two-year dummy variables were considered to
account for between-site and between-year variation, and the remaining independent variables that changed
between years within each transect. The dummy vari- ables were entered at step 0, and the stepwise selection
proceeded from there. Once the final models of this stage were determined, the dummy variables were re-
placed by the variables selected in stage 1, and ‘nested’ models with all the non-dummy variables selected in
the two stages were fitted. Finally, it was checked whether the nested models successfully explained the
between-transect and between-year variation by com- paring the models with the dummy variables model
1 with the nested models model 2, using an F test, calculated as: model 1 sum of squares − model 2 sum
of squaresmodel 1 df − model 2 dfmodel 1 resid- ual sum of squaresmodel 1 residual df, the degrees
of freedom being equal to model 1 df − model 2 df, model 1 residual df.
A correlation matrix of the selected variables was generated for the final winter and spring models.
Additionally, significant differences between land-use categories for the variables selected were determined
using one-way ANOVAs for the independent contin- uous variables and chi-square tests for the categorical
ones.
3. Results
3.1. Density estimates In winter, corn bunting density was lowest in inten-
sive, and highest in extensive categories in all three years Table 5, but the differences were not signif-
icant among land-use categories F
2,6
= 2.89, ns, or
among years F
2,6
= 1.83, ns.
In the breeding season, lowest densities occurred again in intensively managed farmland. Montado
supported higher breeding densities than extensive farmland in two years. Differences among land-use
categories were significant F
2,6
= 11.89, p 0.01,
whereas those among years were not F
2,6
= 0.16, ns.
3.2. Environmental models Table
6 presents
the environmental
models. Both nested models successfully explained the
between-transect and
between-year variation
in corn bunting abundance, the comparisons with the
two-stage models being non-significant F
104,165
= 1.33, p 0.05 for the winter models; F
114,203
= 1.10,
p 0.05 for the spring ones. Of the variables selected by the models, fal-
low area was significantly smaller in the inten- sive land-use category than in the others ANOVA
F
2,270
= 98.71, p 0.001, and temperature was
higher in the extensive land-use category than the others F
2,109
= 10.76, p 0.001. The area of oats in
montado was greater than that in the other land-use categories F
2,318
= 4.31, p 0.05. There was no
significant difference in the occurrence of game man- agement among land-use categories χ
2 2
= 0.25, ns.
The variables selected for the winter models were all correlated Fallow versus Intensive: r
271
= − 0.65;
Fallow versus Temperature: r
263
= 0.24; Intensive
versus Temperature: r
334
= − 0.25; p 0.001 for
all, whereas only Oats and Intensive were signifi- cantly correlated in the spring data set r
319
= − 0.12,
p 0.05. 3.3. Invertebrate abundance
Mean numbers of Lepidoptera larvae per sweep net sample ± SE did not differ between extensive
0.29 ± 0.14 and intensive cereal crops 0.23 ± 0.11, t
36
= 0.023, ns, but, across both categories, their
abundance was positively associated with weed cover r
36
= 0.671, p 0.001. There was no such rela-
tionship for Orthoptera r
36
= 0.09, ns, but these
were more abundant in extensive 1.06 ± 0.26 than intensive cereals 0.19 ± 0.09, t
36
= 3.84, p 0.001.
4. Discussion