sd interval bar lengths not comprehensible

[Chr96er]

When running the following code, I don't understand the interval bar lengths when comparing with the linear models summary (line 3). E.g. variable e has the longest bar, but it doesn't have the highest p-value. Am I missing something or is this a bug?

sampleDataset = read.csv("http://pastebin.com/raw/qhge6SDY")
#Variable no. 18 is the dependent variable, all others except row number are independent variables
linear_model <- lm(sampleDataset[,18] ~ ., data=sampleDataset[,2:17])
summary(linear_model)
coefplot(linear_model)

edit: Upon inspecting the results I just noticed, that the lengths seem to be proportional to the std-error, not the p-value. I understand the reasoning in doing this, but I'm wondering whether the latter would be possible (if (!) reasonable) as well. I.e. that the lengths are proportional to the p-value.

[jaredlander]

While computationally that would be easy, basing so much interpretation on p-values is troubling to me. Aside from always being taught that you shouldn't assume a linearity in p-value meaning the recent ASA statement has really put a damper on p-values.

Now I know that confidence intervals are somewhat similar to p-values, they don't carry the same stigma.

[Chr96er]

That makes sense. Thanks for the ASA statement, that was very helpful!