Hi all,
I see some nice graphs where both 90% and 95% CI intervals are plotted (e.g. the line is 95% interval whereas small ticks indicate 90% interval). Could someone point to an example code of this in R? I would really appreciate.
Hi all,
I see some nice graphs where both 90% and 95% CI intervals are plotted (e.g. the line is 95% interval whereas small ticks indicate 90% interval). Could someone point to an example code of this in R? I would really appreciate.
I did that but could not find anything that describes plotting 90% and 95% CIs together. Would appreciate any help.
ask someone else:
Reg y x, 95 90
I assume from the way you stated the problem that you know how to do 95% lines alone?
Replicate that exact same code, but obviously calculating 90% interval values instead of 95, then use the plot function on them on the existing 95% plot but change line type = 2.
what exactly is the problem? get your CIs, plot one set of lines with the 95% CIs, and another set of lines with the 90% CIs. Voila.
Op is good at R, but not Rochester good.
http://tables2graphs.com/doku.php?id=04_regression_coefficients#graphs_that_do_not_appear_in_paper
Is that what you're talking about? If so, code is there.
Since someone asked, my preferred style: https://gist.github.com/4332698
^ so much wasted space in that graph. remember, graphs are supposed to improve on a traditional tabular presentation. sometimes, an ol' fashioned table is the way to go.
^No. That graph reports 18 parameters in a concise fashion that is easily understood. It's not perfect (I don't report the constant in these graphs unless they're substantively important, for instance), but it's better than a table.
My only complaint is that it is in color. Most journals don't accept color graphics except under special circumstances, so I prefer to do things in black and white to start with.
^ i only see 12 estimates reported. and due to differences in scale, the CIs are all over the map. at best, it's ugly. at worst, it's uninformative, since some of the CIs can't even be determined. this is a graph for graphing's sake.
^Okay, I don't actually like THAT graph, but I like the idea. The scaling is a big issue, but that's more a question of implementation than concept.
And in terms of the parameters reported, I can't tell very well, but you could imagine a graph with 3 models, 6 parameters each, color-coded by model, etc. I've used graphs like that and it worked well. But you're right that this is less useful (at best) in actual practice in this case. My point was more conceptual than about that graph in particular.
You must log in to post.