Skip to content

itsadug:: plot_smooth and plot_diff

3 messages · Bert Gunter, Fotis Fotiadis

Fotis Fotiadis
# 
Hi all

I am using bam to analyse the data from my experiment.
It's a learning experiment, "acc" denotes accuracy and "cnd" denotes a
within-subjects variable (with two levels, "label" and "ideo")."Ctrial" is
centered trial (ranging from 1 to 288).

The model is:
bam(acc~ 1 + cnd + s(ctrial) + s(ctrial, sbj, bs = "fs", m = 1), data=data,
family=binomial)

The model doesn't include two different smooths (one for each condition)
since including two smooths does not result to a more parsimonious model,
according to following model comparison:
m0.2: acc ~ 1 + cnd + s(ctrial) + s(ctrial, sbj, bs = "fs", m = 1)

m1.2: acc ~ 1 + cnd + s(ctrial, by = cnd) + s(ctrial, sbj, bs = "fs",
    m = 1)

Chi-square test of fREML scores
-----
  Model    Score Edf Chisq    Df   p.value Sig.
1  m0.2 10183.31   6
2  m1.2 10173.33   8 9.975 2.000 4.654e-05  ***

AIC difference: -2.16, model m0.2 has lower AIC.


So, I'm trying to assess if there's a difference in accuracy between the
two conditions.

When using the plot_smooth function, the model predictions are the ones
shown in Fig.1.
The code used is:
plot_smooth(fm, view="ctrial",
cond=list(cnd="pseudo"),main="Model",xaxt="n",
xlab="Trial",ylab="Proportion Correct", lwd=2, las=2, rm.ranef=TRUE,
rug=FALSE, shade=T, col="red" )
plot_smooth(fm, view="ctrial", cond=list(cnd="ideo"), xaxt="n",
rm.ranef=TRUE, rug=FALSE, shade=T, col="blue", add=T , lty=2, lwd=2)
legend(x=0.8, y=1.5,legend=c('Label', 'Ideogram'),col=c('red', 'blue'),
lty=c(1,2), bty="n", lwd=2)

Since the 95% confidence intervals overlap, I would assume that there is no
difference in accuracy between the two conditions.

I am also using plot_diff to directly plot the difference:
plot_diff(fm, view="ctrial",comp=list(cnd=c("pseudo", "ideo")),
transform.view=dnrmlz,rm.ranef=T)
(dnrmlz is a simple function to de-normalize trial)

The output of the function is:
Summary:
* ctrial : numeric predictor; with 100 values ranging from -1.725936 to
1.725936.
* sbj : factor; set to the value(s): aggmpo96. (Might be canceled as random
effect, check below.)
* NOTE : The following random effects columns are canceled: s(ctrial,sbj)

* Note: x-values are transformed.
            Significant
1 0.759461 - 288.240539

So, it seems that accuracy in the label condition is higher compared to the
ideo condition throughout the experiment.
This result seems to contradict the previous one.

I am obviously misinterpreting something.
Any ideas on what am I doing wrong?

Thank you in advance for your time,
Fotis
Bert Gunter
# 
To be clear, I know nothing about bam; I just wanted to correct a
statistical error:

"Since the 95% confidence intervals overlap, I would assume that there is no
difference in accuracy between the two conditions."

That is false. You need to look at a CI for the difference.

As you appear to be confused about the statistical issues, I suggest
you post on a statistical site like stats.stackexchange.com or consult
a local statistician.

Cheers,
Bert


Bert Gunter

"The trouble with having an open mind is that people keep coming along
and sticking things into it."
-- Opus (aka Berkeley Breathed in his "Bloom County" comic strip )
On Sun, Jun 12, 2016 at 7:03 AM, Fotis Fotiadis <fotisfotiadis at gmail.com> wrote:
Fotis Fotiadis
# 
Dear Bert,
Thank you for your response

Best,
Fotis
On Sun, Jun 12, 2016 at 5:50 PM, Bert Gunter <bgunter.4567 at gmail.com> wrote: