Statistical analyses were carried out using Stata v13.1
(StataCorp, College station, Texas). As results did not
differ between men and women, the analyses were not
stratified by sex. All tests were two-sided. As multiple
observations were available for many participants because of consumption of more than one dinner (i.e. a red
meat or a fish dinner was consumed on both recall days),
mixed models with a random effect (random intercept)
for participant were used to adjust for this dependency in
the data. Because of the large number of non-consumers
for each of the presented side dishes, the continuous side
dish variables were dichotomized into dinners containing/
not containing each of the side dishes, and the results
are presented as percentages of dinners containing each
side dish (Table 2). For the comparison of the percentages
of red meat and fish dinners containing each side dish,
a logistic mixed model was used with using or not using
each of the side dishes as dependent variables. These
models were adjusted for sex, BMI, educational level,
interest in a healthy diet, smoking habits, whether the
day was a normal day with regard to food and beverage intakes, and weekday versus weekend. Because of the
multiple comparisons, the significance level was set to
pB0.003 (pB0.05 divided by 18 tests, one for each side
dish per age group).
For the comparisons of nutrient intakes between the
red meat and fish dinners (Table 3), mixed models with
energy intake and each of the listed nutrients as dependent
variables were used. The intakes of protein, fat, fatty
acids, carbohydrates, and added sugars are presented as
the percentage of energy intake from each nutrient. Fiber
intake is presented as grams per megajoule (MJ). Because
of many non-consumers, alcohol intake was dichotomized
into dinners containing/not containing alcohol, and the
results are presented as percentages of dinners containing
alcohol. The intakes of iron, folate, vitamin C, b-carotene,
and vitamin D are presented as absolute intakes per dinner.
Linear mixed models were used for all analyses except
for alcohol, for which a logistic mixed model was used.
The models were adjusted for sex, age group, BMI,
educational level, interest in a healthy diet, smoking habits,
whether the day was a normal day with regard to food and
beverage intakes, and weekday versus weekend. Because
of the multiple comparisons, the significance level was set
to pB0.003 (pB0.05 divided by 15 tests).
Mixed models were also used for the comparison of
the whole-day intakes of the same nutrients on the days
with red meat versus fish dinners (Table 4). Alcohol intake was dichotomized into consumer/non-consumer as
described above. These models were adjusted for the same
independent variables as those comparing the individual
dinners, and the significance level was set to pB0.003 as
for the previous models.