A question about bayesian credible interval of an interaction fit in brms

Interfaces brms
,
1
plot cno. id canopy leaf_age date area cage
43 RB122-2 RB122 ho old may 67.71 three
19 RB45-2 RB45 ho old may 66.18 three
19 RB45-1 RB45 ho old may 74.53 three
19 E58-1 E58 ho old may 50.05 two
19 E58-1 E58 ho old may 65.81 two
20 E42-2 E42 ho old may 66.22 two
44 D399-2 D399 ho old may 66.11 three
20 E42-1 E42 ho old may 75.73 two
20 D350-2 D350 ho old may 51.36 three
44 D399-1 D399 ho old may 63.35 three
20 RB44-1 RB44 ho old may 49.32 three
20 D350-1 D350 ho old may 44.67 three
20 RB44-2 RB44 ho old may 43.96 three
43 G677-1 G677 ho old may 50.12 three
43 G677-2 G677 ho old may 55.51 three
33 E241-1 E241 bu old may 62.52 two
33 E241-2 E241 bu old may 48.57 two
43 RB122-1 RB122 ho old may 46.65 three
19 G449-2-o G449 ho old oct 126.84 three
19 G445-1-o G445 ho old oct 137.13 three
32 RB109_1 RB109 bu old aug 43.19 two
20 D367-2-o D367 ho old oct 99.05 three
34 G758-1-o G758 bu old oct 59.82 three
43 E207-2 E207 ho old may 59.92 two
34 G777-1 G777 bu old may 34.17 three
19 G460-2-o G460 ho old oct 81.86 three
34 G758-2-o G758 bu old oct 53.62 three
19 G457_1 G457 ho old aug 82.60 three
43 E207-1 E207 ho old may 53.33 two
19 RB45-1-o RB45 ho old oct 109.71 two
19 RB52_1 RB52 ho old aug 67.39 two
19 G449-1-o G449 ho old oct 114.96 three
43 RB22-1-o RB22 ho old oct 74.42 two
19 G460-1-o G460 ho old oct 100.64 three
34 G777-2 G777 bu old may 42.58 three
19 G445-2-o G445 ho old oct 111.50 three
33 RB113-2 RB113 bu old may 54.02 three
34 711_1 711 bu old aug 67.09 three
19 760_1 760 ho old aug 76.67 three
33 E247-1 E247 bu old may 26.74 two
33 E237-2 E237 bu old may 61.20 two
33 E885-2 E885 bu old may 28.82 three
34 711_2 711 bu old aug 67.07 three
19 760_2 760 ho old aug 83.98 three
20 D350-2-o D350-2 ho old oct 52.69 three
43 RB22-2-o RB22 ho old oct 78.49 two
19 G460_1 G460 ho old aug 76.92 three
19 RB45-2-o RB45 ho old oct 87.26 two
43 677_2 677 ho old aug 35.29 three
33 E885-1 E885 bu old may 31.45 three
43 677_1 677 ho old aug 49.66 three
33 E237-1 E237 bu old may 51.19 two
34 G772-1-o G772 bu old oct 49.23 three
19 G460-2-n G460 ho new oct 42.01 three
34 G773-1 G773 bu old may 39.63 three
34 G777-2-o G777-2 bu old oct 57.86 three
19 G460_2 G460 ho old aug 79.01 three
20 D367-1-n D367 ho new oct 97.50 three
20 D367-1-o D367 ho old oct 73.40 three
19 RB52_2 RB52 ho old aug 68.07 two
33 E247-2 E247 bu old may 35.17 two
20 D362-1-o D362 ho old oct 62.49 three
34 G774_1 G774 bu old aug 56.49 three
33 RB113-1 RB113 bu old may 43.27 three
20 RB43-2-o RB43 ho old oct 56.73 two
19 G457_2 G457 ho old aug 71.22 three
34 G773-2 G773 bu old may 32.55 three
32 RB109-1-o RB109 bu old oct 48.13 two
20 D350-2-n D350-2 ho new oct 59.70 three
19 G445-1-n G445 ho new oct 119.05 three
34 G777-1-o G777-1 bu old oct 48.51 three
19 G449-2-n G449 ho new oct 88.12 three
20 RB43-1-o RB43 ho old oct 58.36 two
43 RB123_2 RB123 ho old aug 62.74 two
33 886_1 886 bu old aug 44.47 three
43 RB124_1 RB124 ho old aug 54.47 two
33 G794_1 G794 bu old aug 66.57 three
19 G460-1-n G460 ho new oct 47.99 three
34 G774_2 G774 bu old aug 56.12 three
20 D367-2-n D367 ho new oct 84.80 three
43 G701_2 G701 ho old aug 64.30 three
43 RB124_2 RB124 ho old aug 64.55 two
20 RB43-2-n RB43 ho new oct 84.56 two
19 758_1 758 ho old aug 49.75 three
20 D350-1-n D350-1 ho new oct 58.38 three
32 RB109-2-o RB109 bu old oct 51.28 two
33 886_2 886 bu old aug 46.41 three
43 G701_1 G701 ho old aug 72.90 three
33 RB112-1 RB112 bu old may 17.74 three
10 RB60-1 RB60 bu old may 14.42 three
33 885-2-o 885 bu old oct 31.15 three
19 RB45-2-n RB45 ho new oct 77.90 two
33 RB112-2 RB112 bu old may 15.71 three
33 885-1-o 885 bu old oct 27.88 three
20 RB43-1-n RB43 ho new oct 96.43 two
34 G758-1-n G758 bu new oct 27.38 three
33 G795-1-o G795 bu old oct 23.66 three
19 758_2 758 ho old aug 36.91 three
19 G445-2-n G445 ho new oct 101.83 three
34 706_1 706 bu old aug 56.63 three
33 G794_2 G794 bu old aug 57.94 three
33 G795-2-o G795 bu old oct 22.48 three
33 G792-2-o G792 bu old oct 66.28 three
20 D362-2-n D362 ho new oct 70.47 three
20 D350-1-o D350-1 ho old oct 53.86 three
33 RB110_2 RB110 bu old aug 14.27 two
19 G449-1-n G449 ho new oct 54.05 three
19 RB45-1-n RB45 ho new oct 106.44 two
33 G792-1-o G792 bu old oct 54.77 three
20 D362-2-o D362 ho old oct 55.41 three
10 RB60-2 RB60 bu old may 16.34 three
43 RB22-1-n RB22 ho new oct 86.71 two
19 760_2 760 ho new aug 112.07 three
19 G460_2 G460 ho new aug 131.32 three
34 706_2 706 bu old aug 44.76 three
43 RB123_1 RB123 ho old aug 66.47 two
20 D362-1-n D362 ho new oct 60.21 three
33 G789_1 G789 bu old aug 30.01 three
43 RB22-2-n RB22 ho new oct 79.11 two
33 RB113_2 RB113 bu old aug 38.75 two
19 G460_1 G460 ho new aug 113.27 three
33 RB110_1 RB110 bu old aug 17.49 two
12 RB60-1-n RB60 bu new oct 11.75 two
34 G777-1-n G777-1 bu new oct 66.50 three
32 RB109_2 RB109 bu old aug 49.48 two
34 G777-2-n G777-2 bu new oct 55.00 three
33 G789_2 G789 bu old aug 35.44 three
12 RB62-2-o RB62 bu old oct 15.40 two
19 760_1 760 ho new aug 131.57 three
33 G795-1-n G795 bu new oct 14.40 three
33 885-2-n 885 bu new oct 29.45 three
19 G457_1 G457 ho new aug 122.19 three
43 RB124_1 RB124 ho new aug 20.51 two
12 RB60-1-o RB60 bu old oct 16.25 two
12 RB62-1-n RB62 bu new oct 30.20 two
43 RB124_2 RB124 ho new aug 40.52 two
33 RB113_1 RB113 bu old aug 42.69 two
33 G792-2-n G792 bu new oct 50.32 three
12 RB62-1-o RB62 bu old oct 15.85 two
34 G758-2-n G758 bu new oct 25.58 three
34 G772-2-n G772 bu new oct 33.48 three
33 G792-1-n G792 bu new oct 58.03 three
32 RB109-1-n RB109 bu new oct 32.66 two
34 G772-1-n G772 bu new oct 20.94 three
19 G457_2 G457 ho new aug 111.57 three
32 RB109-2-n RB109 bu new oct 28.16 two
12 RB60-2-o RB60 bu old oct 17.87 two
43 677_1 677 ho new aug 63.57 three
33 885-1-n 885 bu new oct 24.20 three
12 RB62-2-n RB62 bu new oct 23.00 two
43 677_2 677 ho new aug 63.58 three
33 G795-2-n G795 bu new oct 13.00 three
19 RB52_1 RB52 ho new aug 25.93 two
32 RB109_1 RB109 bu new aug 64.87 two
19 758_2 758 ho new aug 35.90 three
19 RB52_2 RB52 ho new aug 28.87 two
34 G774_2 G774 bu new aug 43.90 three
43 G701_2 G701 ho new aug 49.97 three
33 886_1 886 bu new aug 25.78 three
34 G774_1 G774 bu new aug 64.00 three
33 RB110_2 RB110 bu new aug 27.45 two
19 758_1 758 ho new aug 41.25 three
12 RB60-2-n RB60 bu new oct 10.59 two
33 RB110_1 RB110 bu new aug 30.39 two
34 711_2 711 bu new aug 42.04 three
43 G701_1 G701 ho new aug 56.88 three
32 RB109_2 RB109 bu new aug 62.28 two
43 RB123_2 RB123 ho new aug 68.30 two
33 RB113_1 RB113 bu new aug 75.86 two
34 706_2 706 bu new aug 29.05 three
33 G794_1 G794 bu new aug 54.92 three
33 RB113_2 RB113 bu new aug 80.24 two
34 706_1 706 bu new aug 33.34 three
33 G789_1 G789 bu new aug 31.78 three
33 886_2 886 bu new aug 27.30 three
43 RB123_1 RB123 ho new aug 70.25 two
33 G794_2 G794 bu new aug 57.50 three
34 711_1 711 bu new aug 38.25 three
34 G772-2-o G772 bu old oct 57.86 three
33 G789_2 G789 bu new aug 34.07 three

I fitted a mode like this:

fit_area_august<-brm(formula=area~canopy*leaf_age+(canopy+leaf_age|id),
family = gaussian(link=“identity”),
data=data_shape_august,
seed=1,
prior=c(set_prior("",class=“Intercept”),
set_prior("",class=“sigma”)),
chains=4,
iter=50000,
warmup=20000,
thin=1,
control = list(adapt_delta=0.99,max_treedepth = 15,stepsize=0.001))

then results in figure like this:

ask
ask845×774 62.1 KB

figure code:
fit<-conditional_effects(fit_october_area,effect=“leaf_age:canopy”,re_formula = NULL)
est_area_october ← as.data.frame(fit[[1]])
fig_october_area<-ggplot(data=data_shape_october,mapping=aes(x = leaf_age, y = area))+scale_color_grey(start=0, end=0.5)+scale_fill_grey()+ theme_bw()+
geom_jitter(aes(color = canopy), shape=79,alpha = 0.6,size=4,
position = position_jitterdodge(jitter.width = 0.05,jitter.height = 0),
show.legend = FALSE) +geom_pointrange(data=est_area_october,mapping=aes(x=leaf_age,y=estimate__, ymin = lower__, ymax = upper__,
color = canopy),size=1,position = position_dodge(width = 0.75))+
labs(color=“Canopy species”)+
theme(legend.position="",
legend.text=element_text(size=20,family=“serif”,colour = “black”),
legend.title=element_text(size=20,family=“serif”,colour = “black”),
plot.title = element_text(size = 20,family=“serif”,colour = “black”),
text=element_text(size=20,family=“serif”,colour = “black”),
axis.text = element_text(size = 20, family=“serif”,colour = “black”),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
panel.background = element_blank(),
strip.background = element_rect(fill = “lightblue”),
strip.text = element_text(size = 20))+
coord_cartesian(ylim = c(0, 150))+
labs(x=“Leaf age”,y=expression(“Leaf area (”~cm^2~")"),title="")

My quesions:

  1. Why 95% credible interval outside the observed values, was my model wrong?

  2. In my understanding, the random term (canopy+leaf_age|id) means the random effect id is affected by canopy and leaf_age, and the effects showed in my figure when I uesd re_formula = NULL in conditional_effects(), Is my understanding correct?

Thank you very much!

2

help me ~

3

First thing would be to plot your raw data. Can you drop a plot of that in here?

Second,
The thing that stands out to me is that your iters, warmup, and adapt delta are way to high. I would run the model with 2000 iters and whatever the default warmup is (500?). Same with adapt_delta, max_treedepth, and stepsize. If you get errors with those as the default that tells you something is likely wrong with your data or model.

5

leaf area data.csv (7.0 KB)

This is my row data, do you have other better models for my data? I want to compare the area under two canopy species (bu and ho)

and I think you are quite right, I will test the model in a low setting.

1 Like
6

Oh and you don’t need to set the seed value in them model right now either.

8

I see, thank you!