Format markers and x-label on scatterplot

[Scott Rick]

Hi. The code for my twoway scatterplot generates the graph below. I want to make the following changes to the graph:

1. change the markers to white with black outline
2. have the X-axis label show the years (2002, 2003, 2004) with smaller markers for the months in between each year marker
3. remove the legend

Any help will be much appreciated

set scheme s1color
 
twoway (scatter one time), title("Teleconsults trend") subtitle("(2022-2024)") ///
ytitle("Number of teleconsults") yscale(range(0 .)) ylabel(#6, labsize(small) ///
angle(horizontal) nogrid) ///
 xtitle(Year) graphregion(fcolor(white)) || lfit one time


* Example generated by -dataex-. To install: ssc install dataex
clear
input float(year month) long one float time
2022  1 2233  1
2022  2 2224  2
2022  3 2326  3
2022  4 2052  4
2022  5 2059  5
2022  6 1656  6
2022  7 1574  7
2022  8 1386  8
2022  9 1361  9
2022 10 1071 10
2022 11 1154 11
2022 12 1109 12
2023  1 1082 13
2023  2 1435 14
2023  3 1447 15
2023  4 1172 16
2023  5 1176 17
2023  6 1209 18
2023  7 1289 19
2023  8 1759 20
2023  9 1480 21
2023 10 1311 22
2023 11 1186 23
2023 12 1208 24
2024  1 1345 25
2024  2 1277 26
2024  3 1185 27
2024  4 1257 28
2024  5 1209 29
2024  6 1405 30
2024  7 1491 31

[Nick Cox]

1 and 3 are easy details well documented.

2 is more challenging. I've not done quite what I think you're asking, but you can adapt this. See for the main idea

https://www.stata-journal.com/articl...article=gr0030

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(year month) long one float time
2022  1 2233  1
2022  2 2224  2
2022  3 2326  3
2022  4 2052  4
2022  5 2059  5
2022  6 1656  6
2022  7 1574  7
2022  8 1386  8
2022  9 1361  9
2022 10 1071 10
2022 11 1154 11
2022 12 1109 12
2023  1 1082 13
2023  2 1435 14
2023  3 1447 15
2023  4 1172 16
2023  5 1176 17
2023  6 1209 18
2023  7 1289 19
2023  8 1759 20
2023  9 1480 21
2023 10 1311 22
2023 11 1186 23
2023 12 1208 24
2024  1 1345 25
2024  2 1277 26
2024  3 1185 27
2024  4 1257 28
2024  5 1209 29
2024  6 1405 30
2024  7 1491 31
end 

gen mdate = ym(year, month)

set scheme s1color

scatter one mdate, title("Teleconsults trend") subtitle("(2022-2024)") ///
ytitle("Number of teleconsults") yscale(range(0 .)) ylabel(#6, labsize(small) ///
angle(horizontal) nogrid) ms(Oh) mc(black) ///
xtick(`=ym(2021, 12) + 0.5' `=ym(2022,12) + 0.5' `=ym(2023, 12) + 0.5',  tlength(*3)) ///
xla(`=ym(2022, 6) + 0.5' "2022"  `=ym(2023, 6) + 0.5' "2023" `=ym(2024, 4)' "2024", noticks) ///
 xtitle("") graphregion(fcolor(white)) || lfit one mdate, legend(off) xli(`=ym(2022,12) + 0.5' `=ym(2023, 12) + 0.5', lc(gs12))

[Scott Rick]

Nick Cox : Thanks a lot. That worked very well. If I may - could you please explain this bit?

  
 xtick(`=ym(2021, 12) + 0.5' `=ym(2022,12) + 0.5' `=ym(2023, 12) + 0.5',  tlength(*3)) /// xla(`=ym(2022, 6) + 0.5' "2022"  `=ym(2023, 6) + 0.5' "2023" `=ym(2024, 4)' "2024", noticks) ///

I'm adding on more data: 6 months for 2020 and 12 months for 2021, and if I use the same code I get the graph below. I'm trying to figure how to modify the code to allow for this additional data and understanding it would be very helpful. Thanks again!

Updated data:


input float(year month) long one
2020  7 1585
2020  8 2460
2020  9 2778
2020 10 3284
2020 11 3280
2020 12 3262
2021  1 2688
2021  2 2849
2021  3 2806
2021  4 2517
2021  5 4012
2021  6 2665
2021  7 2556
2021  8 2618
2021  9 2693
2021 10 2247
2021 11 2028
2021 12 2217
2022  1 2233
2022  2 2224
2022  3 2326
2022  4 2052
2022  5 2059
2022  6 1656
2022  7 1574
2022  8 1386
2022  9 1361
2022 10 1071
2022 11 1154
2022 12 1109
2023  1 1082
2023  2 1435
2023  3 1447
2023  4 1172
2023  5 1176
2023  6 1209
2023  7 1289
2023  8 1759
2023  9 1480
2023 10 1311
2023 11 1186
2023 12 1208
2024  1 1344
2024  2 1277
2024  3 1185
2024  4 1257
2024  5 1209
2024  6 1405
2024  7 1490

[Nick Cox]

There is nothing subtle here -- just fiddly calculations to work out where to put the labels and the ticks. Did you read the linked paper in #2?

The year labels go in the middle of each year's observations, at a particular month if the number of observations is odd or between two months otherwise.

The ticks go at the end of each year's observations -- so halfway between the December and the January markers -- and I add lines at the same positions.

clear 

input float(year month) long one
2020  7 1585
2020  8 2460
2020  9 2778
2020 10 3284
2020 11 3280
2020 12 3262
2021  1 2688
2021  2 2849
2021  3 2806
2021  4 2517
2021  5 4012
2021  6 2665
2021  7 2556
2021  8 2618
2021  9 2693
2021 10 2247
2021 11 2028
2021 12 2217
2022  1 2233
2022  2 2224
2022  3 2326
2022  4 2052
2022  5 2059
2022  6 1656
2022  7 1574
2022  8 1386
2022  9 1361
2022 10 1071
2022 11 1154
2022 12 1109
2023  1 1082
2023  2 1435
2023  3 1447
2023  4 1172
2023  5 1176
2023  6 1209
2023  7 1289
2023  8 1759
2023  9 1480
2023 10 1311
2023 11 1186
2023 12 1208
2024  1 1344
2024  2 1277
2024  3 1185
2024  4 1257
2024  5 1209
2024  6 1405
2024  7 1490
end 


gen mdate = ym(year, month)

set scheme s1color

* end year ticks 
forval y = 2020/2023 {
    local tickpos = ym(`y', 12) + 0.5 
    local tickcall  `tickcall' `tickpos'
}

* year labels 
local labelpos = ym(2020, 9) + 0.5 
local labelcall `labelcall' `labelpos' "2020"

forval y = 2021/2023 { 
    local labelpos = ym(`y', 6) + 0.5 
    local labelcall `labelcall' `labelpos' "`y'"
}

local labelpos = ym(2024, 4)
local labelcall `labelcall' `labelpos' "2024"

mac li 

scatter one mdate, title("Teleconsults trend") subtitle("(2020-2024)") ///
ytitle("Number of teleconsults") yscale(range(0 .)) ylabel(#6, labsize(small) ///
angle(horizontal) nogrid) ms(Oh) mc(black) ///
xtick(`tickcall',  tlength(*3)) ///
xla(`labelcall', noticks) ///
 xtitle("") graphregion(fcolor(white)) || lfit one mdate, legend(off) xli(`tickcall', lc(gs12))

Here are the relevant local macros:

_labelcall:     728.5 "2020" 737.5 "2021" 749.5 "2022" 761.5 "2023" 771 "2024"

_tickcall:      731.5 743.5 755.5 767.5

Sometimes it's easier just to work them out by hand and with mental arithmetic.

PS The trend doesn't look very linear to me.

[Nick Cox]

My impression is that many Stata users underestimate the scope for the display command to do little calculations on the fly. So, I would start with

* What's halfway between June and July 2021?

. di ym(2021, 6) + 0.5
737.5

* 7 months' data for 2024, so plot at April 2024?

. di ym(2024, 4)
771

Here's a way to do it otherwise: Loop over the years from first to last, and find the mean monthly date for each year and the last monthly date for each year except the last.

clear 

input float(year month) long one
2020  7 1585
2020  8 2460
2020  9 2778
2020 10 3284
2020 11 3280
2020 12 3262
2021  1 2688
2021  2 2849
2021  3 2806
2021  4 2517
2021  5 4012
2021  6 2665
2021  7 2556
2021  8 2618
2021  9 2693
2021 10 2247
2021 11 2028
2021 12 2217
2022  1 2233
2022  2 2224
2022  3 2326
2022  4 2052
2022  5 2059
2022  6 1656
2022  7 1574
2022  8 1386
2022  9 1361
2022 10 1071
2022 11 1154
2022 12 1109
2023  1 1082
2023  2 1435
2023  3 1447
2023  4 1172
2023  5 1176
2023  6 1209
2023  7 1289
2023  8 1759
2023  9 1480
2023 10 1311
2023 11 1186
2023 12 1208
2024  1 1344
2024  2 1277
2024  3 1185
2024  4 1257
2024  5 1209
2024  6 1405
2024  7 1490
end 


gen mdate = ym(year, month)

set scheme s1color

su year, meanonly 
local first = r(min)
local last = r(max)

forval y = `first'/`last' { 
    su mdate if year == `y', meanonly 
    local labelcall `labelcall' `r(mean)' "`y'"
    local tickpos = r(max) + 0.5 
    if `y' < `last' local tickcall `tickcall' `tickpos'
}

mac li 

scatter one mdate, title("Teleconsults trend") subtitle("(2020-2024)") ///
ytitle("Number of teleconsults") yscale(range(0 .)) ylabel(#6, labsize(small) ///
angle(horizontal) nogrid) ms(Oh) mc(black) ///
xtick(`tickcall',  tlength(*3)) ///
xla(`labelcall', noticks) ///
 xtitle("") graphregion(fcolor(white)) || lfit one mdate, legend(off) xli(`tickcall', lc(gs12))

[Scott Rick]

Thanks a lot Nick Cox. You are right, it didn't strike me to use the display command!

[Nick Cox]

Meanwhile, I pushed your data through localp from SSC, which is just a wrapper for lpoly.

https://www.statalist.org/forums/for...ial-regression

The code is the same as my last except for

localp one mdate, title("Teleconsults trend") subtitle("(2020-2024)") ///
ytitle("Number of teleconsults") yscale(range(0 .)) ylabel(#6, labsize(small) ///
angle(horizontal) nogrid) ms(Oh) mc(black) ///
xtick(`tickcall',  tlength(*3)) ///
xla(`labelcall', noticks) ///
 xtitle("") graphregion(fcolor(white)) xli(`tickcall', lc(gs12))

There is always scope for different taste and judgment, but this result supports any impression from eyeballing the data that the curve is levelling off.

[Scott Rick]

Nick Cox Thanks, Nick, this is great! One quick question - I ran your code without the title, since I don't want it. I now get the graph below with the R² and RMSE in place of the title. Is there anyway for me to remove that, as well as the kernel/degree details at the bottom of the graph?

[Nick Cox]

localp is just a wrapper for lpoly. In addition to a title() it is making use of subtitle() note().

My main point as programmer is that at best what localp shows is one smooth, specified by particular choices. Even if you or your readers don't care how that was done, it's a retrograde step just to present it as a smooth curve.