Published online by Cambridge University Press: 22 January 2016
This article takes a fresh look at the long-running debate on whether the Unionist party owed its electoral success between the Third Reform Act and the Great War predominantly to ‘negative’ factors: principally, low turnout; poor Liberal organization; and a reliable and consistent middle-class vote. Taking advantage of recently digitized election datasets, it conducts the most extensive statistical study thus far attempted, to argue that recent revisionist historians have dismissed too readily the traditional ‘negative Unionism’ thesis associated with J. P. Cornford. It conducts an extensive analysis of the relationship between turnout and Unionist support on national, constituency, and regional levels, and finds that the much-disputed traditional interpretation that Conservatives benefited from low polls in the late Victorian period is broadly borne out in England. Additionally, this article also investigates the wider impact of uncontested constituencies in this period, arguing that the large number of seats left unfought by the Liberals was even more electorally grievous than the raw numbers imply. Both these findings suggest that the Unionists benefited from a still more substantial structural advantage in the late Victorian period than historians have previously assumed. While important aspects of Unionist language and strategy were undoubtedly positive, they were nonetheless underpinned by negative electoral foundations.
The authors are grateful to Paul Readman, Frankie Evans, and especially David Martin, for comments on earlier drafts, and to Iain Sharpe for suggesting references.
1 J. France, ‘Salisbury and the Unionist alliance’, in R. Blake and H. Cecil, eds., Salisbury: the man and his policies (Basingstoke, 1987), pp. 220–34; R. Shannon, The age of Salisbury, 1881–1902: Unionism and empire (London, 1996), pp. 171, 182; M. Bentley, Lord Salisbury's world: Conservative environments in late Victorian Britain (Cambridge, 2004), p. 157.
2 This was political commentator and historian Sir Henry Maine's description of the new system. Another historian, Lord Acton, described the new government elected in 1885 as ‘the first of our democratic constitution’. See H. Maine, Popular government (London, 1886), p. 92. Lord Acton cited in T. A. Jenkins, The Liberal ascendancy, 1830–1886 (Basingstoke, 1994), p. 208.
3 Cornford, J. P., ‘The transformation of Conservatism in the late nineteenth century’, Victorian Studies, 7 (1963), pp. 35–66Google Scholar; ‘The adoption of mass organization by the British Conservative party’, in E. Allardt and Y. Littunen, eds., Cleavages, ideologies and party systems, Transactions of the Westermarck Society, x (Helsinki, 1964), pp. 400–24; ‘Aggregate election data and British party alignment, 1885–1910’, in E. Allardt and S. Rokkan, eds., Mass politics: studies in political sociology (New York, NY, 1970), pp. 107–16.
4 Low, S., ‘The rise of the suburbs’, Contemporary Review, 60 (1891), p. 548Google Scholar. Low predicted that ‘The Englishman of the future will be a suburb-dweller. The majority of people of this island will live in the suburbs.’
5 Cornford, ‘Transformation’, p. 52.
7 The Pearson's correlation coefficient quantifies the relationship between two variables, expressing a value between 1 and −1. A positive reading denotes a tendency of the two variables to rise or fall in tandem; a negative reading denotes a tendency of one to rise as the other falls. A reading near 0 indicates a negligible statistical relationship.
8 Ibid., p. 55. The trend here is far from uniform – it applies very strongly to metropolitan seats, and to a lesser extent to Lancashire and Cheshire, and West, East, and South Midlands; it does not apply to Yorkshire and the north. Similarly, it applies far more to boroughs with more than 250,000 inhabitants than it does to those with 50,000–100,000 or 100,000–250,000 inhabitants.
10 Cornford, ‘Aggregate election data’, p. 111.
11 For a different psephological approach that comes to a similar conclusion, see Dunbabin, J. P. D., ‘Parliamentary elections in Great Britain, 1868–1900: a psephological note’, English Historical Review, 81 (1966), pp. 82–99CrossRefGoogle Scholar; for an examination of Lord Salisbury's personal temperament as a backdrop to the ascendancy of negative Unionism, see P. Marsh, The discipline of popular government: Lord Salisbury's domestic statecraft, 1881–1902 (Hassocks, 1978), pp. 9–17 and 301–3; for an explanation of the role of Conservative organization, especially with regard to registers, see ibid., ch. 6, esp. pp. 195–6; and for an analysis of the importance of Liberal abstention and disorganization, and the reversal of these factors in 1906, see N. Blewett, The peers, the parties and the people: the general elections of 1910 (Bristol, 1972), pp. 20–3, 36–42.
12 R. Harris, The Conservatives: a history (Bungay, 2011), p. 199; E. H. H. Green, The crisis of Conservatism (London, 1995), p. 126.
13 Shannon, Age of Salisbury, p. 313; J. Ramsden, An appetite for power: a history of the Conservative party since 1830 (London, 1998), p. 147.
14 B. Evans and A. Taylor, From Salisbury to Major: continuity and change in Conservative politics (Manchester, 1996), p. 12.
15 Marsh, Discipline, p. 195.
16 M. Pugh, The making of modern British politics, 1867–1939 (Oxford, 1993), p. 72. Other notable works which cite low turnout and tight registers as the keystone of pre-1914 Conservative success are R. F. Haggard, The persistence of Victorian Liberalism (Westport, CT, 2001), p. 131; M. Pearce, British political history, 1867–2001: democracy and decline (London, 2002), p. 98; C. Burness, ‘The making of Scottish Unionism, 1886–1914’, in S. Ball and R. Holliday, eds., Mass Conservatism: the Conservatives and the public since the 1880s (London, 2002), pp. 24–5.
18 M. Roberts, Political movements in urban England, 1832–1914 (New York, NY, 2009), p. 116.
22 For examples of these approaches, see esp. K. D. Wald, Crosses on the ballot: patterns of British voter alignment since 1885 (Princeton, NJ, 1983), pp. 82–94; W. Miller, Electoral dynamics in Britain since 1918 (London, 1977).
23 Lawrence and Elliott, ‘Parliamentary election results’, pp. 21–3.
27 Readman, ‘The 1894 general election’, pp. 488–9.
28 K. S. Rix, ‘The party agent and English electoral culture, 1880–1906’ (D.Phil. thesis, Cambridge, 2001), pp. 196–202.
31 Ibid., p. 231. The Pearson coefficients are: Central, −0.5; East, −0.3; North, −0.9; South, 0.4; West, 0.8.
32 M. Roberts, ‘“A terrific outburst of political meteorology”: by-elections and the Unionist electoral ascendency in late Victorian England’, in T. G. Otte and P. Readman, eds., By-elections in British politics, 1832–1914 (Woodbridge, 2013), pp. 177–200. This work is also notable for being the only one discussed here that treats counties on an equal footing with boroughs.
33 Ibid., pp. 197–8. Overall, the readings are −0.26 in boroughs and −0.29 in counties. In the former, Liberal Unionists have a stronger correlation than Conservatives do; in the latter, theirs is weaker.
34 Ibid., p. 198. ‘Villadom’ sees a stronger negative correlation (−0.45), agricultural divisions a weaker one (−0.26), and working-class boroughs the weakest of all (−0.10).
35 A. Windscheffel, Popular Conservatism in imperial London, 1868–1900 (Woodbridge, 2007), pp. 215–24.
36 Twenty-six of fifty-three frequently contested London seats have correlations stronger than −0.8 from 1885 to 1900.
38 Chronologically for 1886–1900, the readings are −0.612, −0.397, −0.067 and −0.318.
39 Windscheffel, Popular Conservatism, pp. 223–4.
40 Ibid., p. 218. A ‘safe’ seat is classed as one which never changed hands between 1885 and 1900 – there were twenty-six such Conservative seats and four Liberal. By 1895, the safe Tory average turnout was always higher than the safe Liberal; in 1900, it was 6.72 percentage points lower.
41 Only 1892 and 1895 are tested – the correlations are 0.640 and 0.363 (0.474 and 0.216 when Holborn, Marylebone, and Westminster are excluded).
42 For instance, Windscheffel finds strong negative correlations throughout the period between turnout and swing in London (see above, n. 38). Furthermore, Roberts finds that the middle-class Leeds constituencies show very strong negative correlations between vote share and turnout (see above, n. 31), while his by-elections analysis indicates that the mild overall negative correlation between these two variables is strongest of all in ‘Villadom’ (see above, n. 34).
43 This article focuses on turnout and Unionist support because this debate undoubtedly lies at the epicentre of the historiographical clash. Given the constraints of space, it does not assess two other variables which might also be used to measure the validity of ‘negative Unionism’ – the size of the electorate, or rateable value per capita. Both groups of data also have very serious drawbacks. In the case of the former, there are so many factors influencing the size of electorate of any constituency that attempting to draw causal links with Unionist vote seems a dangerously arbitrary exercise. In the case of the latter, it is close to impossible to find constituency-level data across the country, forcing reliance on unwieldy ‘surrogate units’ which such data can be mapped to. Such ecological analyses have rightly been criticized for breaking the constituency link which are arguably fundamental to any psephological analysis in this period. See Lawrence and Elliot, ‘Parliamentary election results’, pp. 19–22. All of the statistical analyses in this article are based on relationships at constituency level, and where aggregates exist, these are simply averages taken from groups of constituencies.
44 W. Field, British electoral data, 1885–1949 [computer file], Colchester, Essex: UK Data Archive [distributor], Nov. 2007. SN: 5673, http://dx.doi.org/10.5255/UKDA-SN-5673-1. The original data is taken from F. W. S. Craig, British parliamentary election results, 1885–1918 (Aldershot, 1989), and British parliamentary election results, 1918–1949 (Chichester, 1983).
45 Statistical significance – or the P value – is the probability that an observed correlation exists only by random chance: the ‘null hypothesis’. We use the standard benchmark for significance, which is a P value of 0.05 (a ⩽5 per cent chance that the null hypothesis is true, denoted by an *) and a P of 0.01 for super-significance (a ⩽1 per cent chance, denoted by **). They are included mainly to denote our level of confidence that, in a given group of constituencies (e.g. South-East region in 1900, all seats in 1892, etc.), the observed trends from contested seats would have applied if all seats had actually been contested. Significance testing is also used, in Figures 4.1 and 4.2, to place a confidence margin on constituency-specific correlations, so we can estimate – if there had been further hypothetical elections in this period – the probability that the trend established from the eight that did happen would have held.
46 For this (and all other analyses in this article), seats were only included if contested unambiguously by the party in question, thus excluding (for example) independent Liberals and Conservatives, and crofter candidates. ‘Lib/Lab’ MPs noted by Craig were (as in other election analyses) counted as Liberals. For double-member boroughs where a party had two candidates, their poll was averaged. In the small number of instances where a party contested only one of the two seats, we assumed the presence of a second candidate who polled identically to the first. It should also be noted that many figures for double-member boroughs in the British electoral data spreadsheet are in fact incorrect, and required manual correction.
47 Windscheffel, Popular conservatism, pp. 216–18; Roberts, ‘“Political meteorology”’, p. 197. The textbook in question is A. Bryman and D. Cramer, Quantitative data analysis for social scientists (London, 1994), pp. 170–1.
48 M. K. Le Roy, Research methods in political science (Boston, MA, 2013), p. 126.
49 S. L. Weinberg and S. K. Abramowitz, Data analysis for the behavioural sciences using SPSS (New York, NY, 2002), p. 136; S. Jackson, Research methods and statistics: a critical thinking approach (Belmont, CA, 2012), p. 149; R. P. Di Fabio, Essentials of rehabilitation research: a statistical guide to clinical practice (Philidelphia, PA, 2013), p. 94; D. Baucum, Psychology (New York, NY, 1999), p. 33.
50 T. C. Urdan, Statistics in plain English (Hove, 2010), pp. 81–2.
51 Cornford, ‘Transformation’, p. 55.
52 This is despite the inherent difficult of passing a t-test with a small sample (in this case eight or less) as even a single outlier opposing the general trend is liable to make it fail. de Winter, J. C. F., ‘Using the student's t-test with extremely small sample sizes’, Practical Assessment, Research and Evaluation, 18 (2013), pp. 1–11Google Scholar.
53 Roberts, ‘“Villa Toryism”’, p. 231. He suggests that the negative correlations for the Leeds Central and North constituencies are offset by the positive correlations for the city's other seats, especially Leeds West. But, while this may have been true in Leeds, Figures 4.1 and 4.2 show that, nationwide, there were very few Leeds Wests to offset very many Leeds Norths.
54 The 100 significantly negatively correlated constituencies (all with coefficients of −0.65 or less) are, in descending order of negativity: Oswestry; Thirsk & Malton; Arfon; Lewisham; Chester-le-Spring; Dundee; Dulwich; Hampstead; Manchester North-West; Wigtownshire; Birmingham Bordesley; Ross (Herefordshire); Enfield; Monmouthshire West; Durham; South Molton; St Andrews District; Norwood; Honiton; Bassetlaw; Wandsworth; Durham South-East; Brixton; Aston Manor; Exeter; Droitwich; Partick; Kingston-upon-Hull East; Liverpool West Derby; Manchester South-East; Salford North; Horncastle; St Pancras East; Walthamstow; Govan; St Pancras South; Bristol North; Hyde; Bodmin; Wellington (Shropshire); Whitby; Monmouthshire South; Caernarvon District; St Pancras West; Limehouse; Tynemouth; Stretford; Pembroke & Haverford District; Greenock; Hammersmith; Islington South; Bow & Bromley; Huddersfield; Chesterton; Truro; Romford; Blackburn; Staffordshire North-West; Dumfriesshire; Kennington; Barnstaple; Darwen; Harwich; Montrose District; Deptford; Islington East; Islington West; Leigh; Lanarkshire Mid; Woolwich; St Helens; Bradford West; Leek; Manchester East; Liverpool East Toxteth; St Ives; Rossendale; Sheffield Ecclesall; Holderness; Glasgow Camlachie; Bristol West; Gravesend; Inverness District; Ramsey; Newton; Somerset North; Halifax; Hackney Central; St Pancras North; Salford West; West Ham North; Islington North; Mile End; Bristol South; Clapham; Hackney North; Marylebone East; Rotherhithe; St George; and Tottenham. Comparing this list to Britain as a whole, a number of trends are apparent. One is that London seats comprise 27 per cent of the list, as opposed to 11 per cent of those in the country, while the West Midlands and the South-East are slightly under-represented, and Wessex and the Central Region have no negative seats at all (although this could be a statistical fluke, given the small sizes of both regions). In terms of social cleavage, all urban seats are mildly over-represented, regardless of class, whereas the mixed urban/rural seats are considerably under-represented. However, despite these trends, the group features multiple seats from the vast majority of Pelling's fifteen regions, and all of Blewett's six social cleavage categories. are also all well represented. See H. Pelling, Social geography of British elections, 1885–1910 (London, 1967), pp. xi–xxxi; Blewett, Peers, pp. 488–94.
56 Categorizing Villadom with sufficient precision for quantitative analysis is extremely difficult – though its affluent and professional middle-class nature makes it conducive to the traditional simple (if crude) measure of resident domestic servant per household, which is the basis for Blewett's categories of ‘Urban Middle Class’ (A), ‘Urban Mixed Class’ (B), and ‘Urban Working Class’ (C). The substantial overlap between Villadom and category A and B – which are often wealthy constituencies in central and commercial parts of cities – is self-evident. However, as Roberts points out (‘“Political meteorology”’, p. 198 n. 100), an exhaustive list of Villadom could also include suburbanized county divisions, which mainly fall in category D (‘Mixed Urban/Rural’).
57 Scatterplots for these two elections look very similar to those already presented in Figure 6.2, and are omitted in the interests of space.
58 Correlations excluding Scotland, Wales, and Liberal Unionists are −0.341**, −0.23**, −0.292**, and −0.519** for 1886, 1892, 1895, and 1900 respectively. In all but 1895, the negative correlations are thus slightly strengthened. Correlations excluding Scotland, Wales, and English mining seats are −0.336**, −0.29**, −0.474**, and −0.574** for these four respective elections. Interestingly, both 1910 elections register identical −0.223** correlations in this case, leaving only 1885 and 1906 without any trend.
59 R-Squared, or the coefficient of determination, is a statistical measure of how close a fit the data are to our regression line. It is a measure of what proportion of the dependent variable (Unionist vote share) can be accounted for by the independent variable (turnout). Naturally, because turnout is just one of a host of factors which might help predict Unionist vote share (region, class, and previous election results are examples of others), we do not expect its R-Squared value to be very high on its own. We use the R-Squared value not as a universal benchmark of acceptability (something it is often mistakenly used as) but instead to compare our regression models with each other, so we can select the best one. See F. Moksony, ‘Small is beautiful: the use and interpretation of R-Squared in social research’, Szociológiai Szemle, special issue (1990), pp. 130–8.
60 These modifications increase our R-Squared values from 0.01 (1886), 0 (1892), 0.026 (1895), and 0.096 (1900) for linear British trendlines to 0.165, 0.103, 0.253, and 0.319 respectively for polynomial English trendlines.
61 The English turnout in 1886 was 74.4 per cent. An increase of 1.7 percentage points to 76.1 per cent (the level of 1900) results in a swing of 0.29 per cent against the Unionists in England, and a loss of six seats. Given the particular curve of the trendline for this election, this predicted small swing is more or less identical with turnouts above 80 per cent.
62 If the 1892 English turnout were raised to 82.4 per cent, we can predict a 0.67 per cent decline in Unionist vote and the loss of fifteen English seats on uniform swing. For a turnout of 84.3 per cent, the decline is 0.87 per cent, resulting in nineteen seat losses.
63 The only specific studies for this period are the brief and general T. O. Lloyd, ‘Uncontested seats in British general elections, 1852–1910’, Historical Journal, 8 (1965), pp. 260–5 and a section from Rix, ‘Party agent’, pp. 254–6.
64 Despite its obvious importance, it is striking that uncontested seats are given cursory attention by Cornford (‘Transformation’, p. 54 n. 44), and all the revisionist works discussed above are silent on the issue.
65 A. L. Lowell, The government of England, ii (London, 1908), p. 59. This was the classic ‘tactical’ view of investing in fighting elections only when and where there was a good chance of victory. This was challenged in the later Edwardian period, where contesting even hopeless seats was increasingly seen as a mark of party virility. See Lloyd, ‘Uncontested seats’, p. 265; Moore, D. C., ‘The matter of the missing contests: towards a theory of the mid-nineteenth-century British political system’, Albion, 6 (1974), pp. 108–10CrossRefGoogle Scholar; K. M. O. Swaddle, ‘Coping with a mass electorate: a study in the evolutions of constituency electioneering in Britain, with special emphasis on the periods which followed the Reform Acts of 1884 and 1910’ (D.Phil. thesis, Oxford, 1990), p. 247.
66 The only addition to this rule was that, if the party in question left a seat uncontested which they themselves won unopposed in a different election, they are counted as having 100 per cent of the vote in this case. This is unsurprisingly extremely rare. To give a worked example of our approach in Figure 6.1, we will take the Abingdon division of Berkshire, which was uncontested by the Liberals in 1900. In the other seven contests, the Liberals polled 41.3 per cent, 32.9 per cent, 47.6 per cent, 42.6 per cent, 51.1 per cent, 43.9 per cent, and 41.6 per cent. The average Liberal vote – and our estimate for 1886 for this seat – is thus 43 per cent.
67 1885 is excluded from this analysis because only twenty seats were left uncontested by the Conservatives and Liberals combined in this election in Great Britain, thus rendering any trends very unreliable.
68 Using the average national vote shares for each election (shown in Figure 9) to calculate whether parties in respective elections over or underperformed their 1885–1910 average, and adjusting our forecasts for uncontested seats accordingly, makes very little difference to the overall picture. On average, the Liberals remain 6.22 percentage points ahead.
69 To give a worked example, the Liberals left 113 seats uncontested in 1886. Looking at those in the other seven elections (791 different readings) we can then apportion occasions where they got a strong poll or a weak poll. In this case, the Liberals scored 50 per cent+ in 128 instances, and 35 per cent or less in 83. This is 14.2 per cent strong results, 9.2 per cent weak. Of the 113, this equates to 16 strong and 10 weak seats.
70 Across the period, 16.17 per cent of Liberal uncontested seats were strong, and 9.61 per cent weak. For the Unionists, these figures were 8.1 per cent strong and 28.14 per cent weak.
71 Lord Salisbury to R. Churchill, 29 Feb. 1884. See W. S. Churchill, Lord Randolph Churchill, i (London, 1906), p. 314.
72 Annual Register (London, 1895), p. 153. Cited in Rix, ‘Party agent’, p. 255.
73 H. Gladstone to J. M. Paulton, 31 Jan. 1903, British Library, Herbert Gladstone papers, Add. MS 46060, fol. 117.
74 H. Gladstone to R. A. Hudson, 30 Aug. 1900, ibid., Add. MS 46020, fol. 59.
75 Lloyd, T. O., ‘The whip as paymaster: Herbert Gladstone and party organization’, English Historical Review, 89 (1974), pp. 792–3Google Scholar.
76 Liverpool Mercury, 15 July 1895.
77 Daily Gazette for Middlesbrough, 24 July 1895.
78 Dundee Courier, 30 Dec. 1898. For the poor Unionist prospects in 1899, see P. Readman and L. Blaxill, ‘Edwardian by-elections’, in Otte and Readman, eds., By-elections, p. 238.
79 Yorkshire Post and Leeds Intelligencer, 30 Jan. 1892.
80 Western Gazette, 8 July 1892.
81 Western Gazette, 12 July 1895.
82 Coventry Herald, 21 Sept. 1900.
83 Cheshire Observer, 12 Oct. 1895.
84 Leeds Mercury, 22 Sept. 1900.
85 Manchester Courier and Lancashire General Advertiser, 21 Feb. 1900.
86 Leeds Times, 3 July 1886.
87 Author unknown, ‘The next general election: a forecast’, Westminster Review, July 1888, p. 394.
88 Yorkshire Post and Leeds Intelligencer, 11 June 1895.
89 Aberdeen Journal, 26 June 1886.
90 Aberdeen Journal, 5 July 1886.
91 For an examination of Unionist attempts to appropriate Liberal discourses in a safe Liberal seat, see C. Macdonald, ‘Loyalty, tradition and language in the evolution of Scottish Unionism: a case study, Paisley, 1886–1910’, in C. Macdonald, ed., Unionist Scotland, 1800–1997 (Edinburgh, 1998), pp. 57–60.
92 Leeds Mercury, 14 July 1886. Almost all the erosion in the Liberal numerical majority was caused by a 20 percentage point fall in turnout. Perhaps in part this was due to the Liberal Unionist identity of the candidate enabling habitual Liberals to abstain, whereas a Tory would have aroused tribal animosity.
93 Sunderland Daily Echo and Shipping Gazette, 24 Sept. 1900.
94 Lloyd, ‘Whip as paymaster’, pp. 800–1.
96 Rix, ‘Party agent’, p. 256.
97 Saturday Review, 24 Oct. 1891, p. 463.
98 H. Dyer, ‘The future of politics’, Westminster Review, Jan. 1896, p. 6.
99 Gladstone, After thirty years (London, 1928), p. 160.
100 Shannon, Age of Salisbury, p. 415.
101 See n. 12 above.
102 France, ‘Salisbury and the Unionist alliance’, p. 223.
103 J. Lawrence, ‘Political history’, in S. Berger, H. Feldner, and K. Passmore, eds., Writing history: theory and practice (London, 2003), pp. 195–6.
104 F. Craig, British electoral facts, 1832–1980 (Chichester, 1981), pp. 79–84.
Full text views reflects PDF downloads, PDFs sent to Google Drive, Dropbox and Kindle and HTML full text views.