Seasonality graphics

[Emanuele Spina]

Hi,

i need help!

I'm testing seasonality in trend research about a rare disease. I found statistical significance over eight year of analysis (96 time points) by applying a generalized linear model.
Then i calculate monthly mean and performed a graph (cycleplot, with a graphical clear seasonality).. I'd like to obtain a line connecting my monthly means.. but i can't found the right command.

My GLM command: (glm trend i.month c.year cos_1 sin_1 time, family (poisson) link(log) eform)
My Cycleplot command: (cycleplot mean month year, xla(1/12) sy(i) myla(J F M A M J J A S O N D)

mean correspond to the combined trend mean over each month

i attached what i obtained and what i'd like to obtain

Can someone help me?

[Nick Cox]

cycleplot is from the Stata Journal.

Is

line mean month, sort

what you want?

[Emanuele Spina]

Maybe on predicted data, and in a way to make it continuous.. as posted in second image!

[Nick Cox]

Sorry, but I am lost here. If you can post a graph, you must know how to make that graph. Fine, but I don't understand what you're missing.

[Emanuele Spina]

That's not my graph! I took it from the web, and i'd like to do something like that.. but i don't know how!

[Emanuele Spina]

The first one is mine, and i have no idea how to transform it into the second one!

[Nick Cox]

cycleplot can produce the second graph.

[Nick Cox]

I was confused too; sorry about that. I am not clear enough about your data to advise. For example, nothing in your glm call makes it obvious to me why hemisphere predictions differ by a factor of almost 10.

[Emanuele Spina]

Make it simple: how can i produce the second graph (northern southern etc.etc.) with cycleplot? Which options can add the continous line?

Beginning with this command: (cycleplot mean month year, xla(1/12) sy(i) myla(J F M A M J J A S O N D)

[Nick Cox]

It's your first graph that could come out of cycleplot, or so I guess.

[Andrew Musau]

Please make an effort to present a data example using the dataex command in the future. You can produce the second graph using twoway. Here is some technique.

*CREATING DATA EXAMPLE FOR ILLUSTRATION
clear
set obs 10
local months "jan feb mar apr may"
set seed 2019
foreach month of local months{
gen `month'= runiformint(15, 40)
}

*PRESERVE INITIAL DATASET
preserve
*COLLAPSE TO GET MEANS BY MONTH
collapse (mean) jan- may
*DUPLICATE MONTHLY MEANS TWICE AND CREATE OBSERVATION #
expand 3
gen t=_n

*RESHAPE TO LONG LAYOUT
rename (jan-may) month#, addnumber(1)
reshape long month, i(t)
label var month "Mean whatever"
*SORT AND ORDER
sort _j
gen order=_n
*GRAPH USING TWOWAY

tw(line month order if _j==1)(line month order if _j==2)(line month order if _j==3) ///
(line month order if _j==4) (line month order if _j==5, leg(off) xtitle("") ///
xlabel(2 "Jan" 5 "Feb" 8 "Mar" 11 "Apr" 14 "May") scheme(s1mono))

Result:

[Emanuele Spina]

Ok, this graph is what i obtain with cycleplot (see my first graph).

Now i'm searching a command for a connection line (see the second graph i posted, that's NOT mine but an example i found in literature).

Can you help me?

[Igor Paploski]

I guess the confusion is that what you are referring to "first" and "second" graph is inverted. You are making the one that looks like Stata, but want to do something similar to the graph that is subdivided into the Northern and Southern Hemispheres, correct?

Try looking if some sort of line graph with smoothing (lowess) doesn't help you:

clear
input long month float y
 5 45
 4 49
 8 50
 1 47
 9 52
 7 51
 6 46
 2 47
12 50
11 51
10 48
 3 42
end
label values month mon
label def mon 1 "apr", modify
label def mon 2 "aug", modify
label def mon 3 "dec", modify
label def mon 4 "feb", modify
label def mon 5 "jan", modify
label def mon 6 "jul", modify
label def mon 7 "jun", modify
label def mon 8 "mar", modify
label def mon 9 "may", modify
label def mon 10 "nov", modify
label def mon 11 "oct", modify
label def mon 12 "sep", modify



twoway (lowess y month, bwidth(0.5)) (scatter y month)

[Emanuele Spina]

Ok. By correcting your input


clear
input long month float y
5 53
4 51
8 47
1 45
9 50
7 46
6 50
2 49
12 42
11 48
10 51
3 50
end
label values month mon
label def mon 4 "apr", modify
label def mon 8 "aug", modify
label def mon 1 "dec", modify
label def mon 2 "feb", modify
label def mon 4 "jan", modify
label def mon 7 "jul", modify
label def mon 6 "jun", modify
label def mon 3 "mar", modify
label def mon 5 "may", modify
label def mon 11 "nov", modify
label def mon 10 "oct", modify
label def mon 9 "sep", modify



twoway (lowess y month, bwidth(0.5)) (scatter y month)

i obtained this graph, that is much more than previously seen on my pc.

Now, the graph to what i'm inspiring (northern and southern, as you have correctly decoded: D) has a more "harmonic" line. I think that the researchers have shown the predicted line according to their model. There's a way to do it?

[Igor Paploski]

Hi Emanuele,

The question for me is what is it that you are trying to show. If you are trying to create just a figure that looks pretty, with a line that seems more "harmonic", you can play around with settings in the graph twoway graphic user interface. Some alternatives are to change the bandwidth or to specify mean on the command (see below). To obtain further info on what that does, please refer to the help file.

twoway (lowess y month, mean bwidth(0.6)) (scatter y month)
help twoway_lowess

If what you want is to plot what the prediction of your model is (a good way to compare how your model is performing when comparing to your data), you can try using the predict command after your model - help predict, and then plot the prediction of your model as a line in your graph (instead of the lowess smoothing). See a mock model below with the prediction plotted against data.

clear
input long month float y
 1 45
 2 49
 3 50
 4 51
 5 53
 6 50
 7 46
 8 47
 9 50
10 51
11 48
12 42
end
label values month mon
label def mon 1 "jan", modify
label def mon 2 "feb", modify
label def mon 3 "mar", modify
label def mon 4 "apr", modify
label def mon 5 "may", modify
label def mon 6 "jun", modify
label def mon 7 "jul", modify
label def mon 8 "aug", modify
label def mon 9 "sep", modify
label def mon 10 "oct", modify
label def mon 11 "nov", modify
label def mon 12 "dec", modify

gen y2 = y*y
poisson y y2
predict y2hat
twoway (line y2hat month) (scatter y month)