Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Scottish referendum results as a sandbox for graphics

    Scotland voted on 18 September 2014 in a referendum on the question Should Scotland be an independent country?

    Apart from the genuine interest in this question (which many Stata users raised at the recent meetings in Boston and London), the dataset created provides a sandbox for those interested in statistical graphics. Results were reported by councils, that is local authority areas.

    I used http://www.bbc.co.uk/news/events/sco...cides/results: this website provides a stacked or divided bar chart and a useful interactive map.

    What else can we do? Here are the results of a morning's play (local time, naturally) together with the data and a script. For once, no user-written programs were used.

    The simplest useful display of the main results I suggest to be a dot chart as reproducible using graph dot.


    Click image for larger version

Name:	scotref1.png
Views:	1
Size:	27.8 KB
ID:	245678


    Small points to note here are changing the size to accommodate the number of points displayed. I find that the default grid of dotted lines in graph dot doesn't always export well to other software, so I tend to change that to very thin, light grey lines. Blue and red were the colours used by the different campaigns and in any case work well together in graphics.

    In many ways the results cry out for a map and the BBC did a good one and no doubt many others will follow. But the standard choropleth (patch) map based on physical space often works badly when, as is very true of Scotland, the range of areas is great (here, 400x variation from smallest to largest). In any case, the precise detail of council area or coast outlines is not really important to understanding, for all that it may be familiar or interesting otherwise. An old and in many ways standard alternative is some kind of cartogram (http://en.wikipedia.org/wiki/Cartogram) in which area is replaced by some other measure (here most obviously electorate), with more or less faithfulness to contiguity or shape.

    A recent map of the USA in the Economist for September 13 2014 -- article "Gunshot detectors: Calling the shots" -- provoked a much simpler alternative. It is also much cruder, but there you go. The idea is essentially to provide a chessboard approximation to the geography as basis for a map cartoon. (Cartoonography? I am not sure that will fly.) The only rule I followed strictly was to respect separation by sea. US readers (especially Stata friends who underline to me that they dropped geography at age 12 because it clashed with AP Calculus) should think of Alaska and Hawaii. Otherwise what was done was a rough compromise with some respect for each of contiguity, position and compass orientation.

    This allows something to be done with graph twoway:

    Click image for larger version

Name:	scotref2.png
Views:	1
Size:	37.1 KB
ID:	245679

    Scotland, so far as I know, lacks a standardized two-letter abbreviation for local authorities similar to that used in the USA, so I grew my own. (DG is already widely used in its own area, and that may be true elsewhere.) The result is, as said, crude, but it allows you to think about the spatial patterns in the data. You just have to think up some integers for the area coordinates, which for a country with 32 areas is a few minutes' work.

    Here is a variation on the same theme, a kind of graphical table:

    Click image for larger version

Name:	scotref3.png
Views:	1
Size:	15.6 KB
ID:	245680


    Here's all the code, for anyone interested at any level:

    Code:
    clear
    input str25 council str2 ab long electorate      yes     no      area    y       x
    "Aberdeen City" "Ab"    175745  41.39   58.61   186     7       7
    "Aberdeenshire" "As"    206486  39.64   60.36   6313    7       6
    "Angus" "An"    93551   43.68   56.32   2182    6       6
    "Argyll & Bute" "AB"    72002   41.48   58.52   6909    5       4
    "Clackmannanshire"      "Cl"    39972   46.20   53.80   159     5       6
    "Comhairle nan Eilean Siar"     "ES"    22908   46.58   53.42   3071    7       2
    "Dumfries & Galloway"   "DG"    122036  34.33   65.67   6426    1       3
    "Dundee City"   "Du"    118729  57.35   42.65   60      6       7
    "East Ayrshire" "EA"    99664   47.22   52.78   1262    2       2
    "East Dunbartonshire"   "ED"    86836   38.80   61.20   175     4       4
    "East Lothian"  "EL"    81945   38.28   61.72   679     2       6
    "East Renfrewshire"     "ER"    72981   36.81   63.19   174     3       2
    "Edinburgh"     "Ed"    378012  38.90   61.10   264     3       6
    "Falkirk"       "Fa"    122457  46.53   53.47   297     4       5
    "Fife"  "Fi"    302165  44.95   55.05   1325    5       7
    "Glasgow"       "Gl"    486219  53.49   46.51   175     3       3
    "Highland"      "Hi"    190778  47.08   52.92   25659   7       4
    "Inverclyde"    "In"    62481   49.92   50.08   160     4       1
    "Midlothian"    "ML"    69617   43.70   56.30   354     2       5
    "Moray" "Mo"    75170   42.44   57.56   2238    7       5
    "North Ayrshire"        "NA"    113923  48.99   51.01   885     3       1
    "North Lanarkshire"     "NL"    268704  51.07   48.93   470     3       4
    "Orkney Islands"        "Or"    17806   32.80   67.20   990     9       6
    "Perth & Kinross"       "PK"    120015  39.81   60.19   5286    6       5
    "Renfrewshire"  "Re"    134735  47.19   52.81   261     4       2
    "Scottish Borders"      "SB"    95533   33.44   66.56   4732    1       5
    "Shetland Islands"      "Sh"    18516   36.29   63.71   1466    11      6
    "South Ayrshire"        "SA"    94881   42.13   57.87   1222    2       1
    "South Lanarkshire"     "SL"    261157  45.33   54.67   1772    2       3
    "Stirling"      "St"    69033   40.23   59.77   2187    5       5
    "West Dunbartonshire"   "WD"    71109   53.96   46.04   159     4       3
    "West Lothian"  "WL"    138226  44.82   55.18   427     3       5
    end
    
    notes : results from http://www.bbc.co.uk/news/events/scotland-decides/results 19 Sept 2014
    notes : areas from http://en.wikipedia.org/wiki/List_of_Scottish_council_areas_by_area 19 Sept 2014
    label var area "area (sq. km)"
    
    set scheme s1color
    
    separate yes, by(yes < 50) veryshortlabel
    graph dot (asis) yes0 yes1, over(council, sort(yes) descending) ysize(7) linetype(line) lines(lc(gs12) lw(vthin)) exclude0 ytitle(% voting Yes) ysc(alt) marker(1, mcolor(red)) marker(1, mcolor(blue)) legend(off) yli(50, lw(vvthin) lc(gs12))
    
    graph export scotref1.png, replace
    
    more
    
    gen yes2 = 5 * floor(yes/5)
    separate y , by(yes2) veryshortlabel
    scatter y30-y55 x, ms(S ..) msize(*5 ..) mlcolor(gs12 ..) mcolor(red red*0.7 red*0.4 red*0.1 blue*0.1 blue*0.4)  aspect(1.35) yscale(off) xscale(off) plotregion(lc(none)) || scatter y x, ms(none) mla(ab) mlabpos(0) mlabcolor(black) legend(col(1) order(1 "30-" 2 "35-" 3  "40-" 4 "45-" 5 "50-" 6 "55-") pos(3))
    
    graph export scotref2.png, replace
    
    more
    
    gen yes3 = floor(yes)
    scatter y y x, ms(none ..) mla(ab yes3) mlabpos(12 0) mlabsize(*.9 *.9) aspect(1.35) yscale(off) xscale(off) plotregion(lc(none)) legend(off)
    
    graph export scotref3.png, replace
    Last edited by Nick Cox; 19 Sep 2014, 06:38.

  • #2
    Fabulous contribution, Nick. Thank you for the education using this timely data!

    Comment


    • #3
      Thanks!

      The question of electorate size was raised privately. Clearly the outcome of a referendum based on total votes depends on electorate sizes as well as proportions of yes and no votes, and none of the graphics above addresses this. A plot of % yes against electorate size helps to make clear that this is important but not crucial to the outcome here (45% yes, 55% no). I've thought about something like a spine plot with councils as one axis and yes or no as another, but it's hard to see that would not be a mess.

      Comment


      • #4
        Many thanks Nick. Just to mention, the provided URL needs correction.
        Roman

        Comment


        • #5
          Thanks. My extra punctuation is interfering. Should be http://www.bbc.co.uk/news/events/sco...ecides/results

          Comment


          • #6
            Great contribution Nick!
            Regards
            --------------------------------------------------
            Attaullah Shah, PhD.
            Professor of Finance, Institute of Management Sciences Peshawar, Pakistan
            FinTechProfessor.com
            https://asdocx.com
            Check out my asdoc program, which sends outputs to MS Word.
            For more flexibility, consider using asdocx which can send Stata outputs to MS Word, Excel, LaTeX, or HTML.

            Comment


            • #7
              Excellent as usual Nick.
              One of the nicest aspect of the new Statalist format, is that now one can more easily see these beautiful graphics!

              Comment


              • #8
                Dear Nick,
                thanks as always for your enlightening contribution.

                Kind regards,
                Carlo
                Kind regards,
                Carlo
                (Stata 19.0)

                Comment

                Working...
                X