By Colman Cassidy

ABSTRACT

Divorce as a constitutional issue, has played a major role in the shaping of modern Irish politics – focusing attention on the dynamics of the referendum as a mechanism for change.   This essay, which is in two parts, looks at the two referendums on divorce that led to the seminal change in the Constitution, 20 years ago.   The first part reviews the background to government/societal initiatives that called for the legalisation of divorce in Ireland, with particular emphasis on the 1986 referendum.   The second part is more substantive in that it uses statistical techniques (logistic regression) to identify a posteriori as ‘predictions’ the factors that led to a Yes vote for divorce on 24 November 1995.  In doing so it focuses on a number of socio-demographic hypotheses (age, urban/rural, for example), along with other factors such as ‘political alienation’ and ‘party allegiance’, to establish ‘predictable’ explanations for the Yes outcome in the referendum.

Part 1: The 1986 Referendum

The two divorce referendums were held within the time-frame, 1986-95.  The first took place in June1986, at the initiative of the Fine-Gael/Labour coalition government and in response to public opinion polls that indicated strong popular support for a constitutional change in the law that would allow divorce.¹⁾Article 41.3.2 of Bunreacht na hÉireann stated that “No law shall be enacted providing for the grant of a dissolution of marriage”.  Significantly, in 1986 the Catholic church did not forbid its members from voting in favour of the proposed change in the law – despite the fact that divorce is contrary to the church’s teaching.  Catholics were asked to review the church’s teaching on the issue and to vote in accordance with conscience.²⁾cf. Girvin, B., ‘The divorce referendum in the republic, June 1986’, Irish Political Studies, Vol.2, pp.91-96. Also, cf. O’Leary, C. and Hesketh, T., ”The Irish abortion and divorce referendum campaigns,’ Irish Political Studies, Vol.3, pp.43-62.  This is not to suggest, however, that the official Catholic position was any less virulent in its opposition to divorce per se.  Indeed, as with other campaigns on key moral issues such as the 1983 ‘pro-life’ (anti-abortion) referendum, there was strong evidence of direct intervention by the church militant, both lay and clerical.³⁾Gallagher, M., ‘Ireland: the referendum as a conservative device?’ in The Referendum Experience in Europe, London: Macmillan, 1996, p.92.
A significant feature that initially appeared to augur well for the government’s strategy was the fact that the opposition Fianna Fáil (the largest of the Irish political parties, containing within its ranks a strong element of traditional Catholic support) officially opted to remain neutral on the issue.⁴⁾ibid.   This a priori ‘neutral stance’ by Fianna Fáil, turned out, a posteriori, to be more ostensible than real.
In addition, the three main Dublin-based national newspaper groups, The Irish Times, Irish Press and Irish Independent strongly supported the proposed change in the law – as did the relatively influential Cork Examiner (now the Irish Examiner), which is widely read in the urban areas of Cork/Munster.⁵⁾Darcy, R. and Laver, M., ‘Referendum dynamics and the Irish divorce amendment’, Public Opinion Quarterly, Vol.54, No.1, Spring 1990, pp.1-20.
In the event, instead of the landslide support to change the law that had initially been mooted, there was a marked swing in public opinion – in a conservative direction – once the 1986 referendum campaign got underway: in April, two months before the referendum, the opinion polls showed that a majority of the electorate was in favour of liberalisation, but once the campaign started in earnest, a major slide became evident.  The anti-divorce lobby proclaimed that a change in the law meant that marriage breakdown would become widespread in Ireland – and ‘deserted’ wives and families would be impoverished as a result.  Jeremiads were also widely dispersed on the subject of property and inheritance rights, particularly in relation to farmland.

Opinion polls showed that support for a change in divorce law in April, fell from 57% to 40%, just one week before the referendum, on 26 June:

Support fell among all social groups, with the greatest drop being among women (from 58 per cent in April to 31 per cent in June).  Consequently, the last pre-vote poll found that men actually favoured legalising divorce, albeit by a small margin, whereas women were opposed by almost a two to one margin.  This gender gap testified to the effectivenes of the anti-divorce campaign, which had sought to instil doubts among women voters in particular.

⁶⁾Gallagher, pp.96-97

As for the outcome of the referendum, on a 62.1% turnout (relatively high for a referendum), divorce was rejected by 36.3% to 63.1%.  What was particularly striking about this result, Darcy and Laver noted, was that such “a short campaign, lasting less than three months, could produce such a massive shift in opinion on an issue that had been discussed and debated for over a decade”.⁷⁾cf. Darcy and Laver, p.2.  Incidentally, the 1986 and 1995 ‘divorce’ referendums were the only ones since 1972 “to induce a turnout of over 60%”. cf. Gallagher, p.102.
To put this in context, surveys carried out by the Market Research Bureau of Ireland, MRBI, clearly illustrate the shift in public opinion on the divorce issue that had taken place in Ireland over the previous 13 years.
Particularly noteworthy is the post-referendum swing back in favour of divorce in the survey conducted in October 1986.  So why were the polls confounded?  In their article on ‘Referendum Dynamics’, Darcy and Laver examine a theoretical model of elite withdrawal and community conflict that appears to go some way to explain what happened in the first Irish referendum on divorce:

An established group or coalition of organisations becomes concerned over an issue.  After a process of internal dialogue, they negotiate with established political elites, a process which may result in the proposal being submitted by those elites for legislation or public ratification…   Once on the agenda, however, the proposal draws new groups into the debate, often ad hoc groups, supported by organisations and networks based outside the established political elite.  Participants tend to be of lower social and political status than established political elites and tend to be motivated by very strongly held ideological beliefs, to which the proposed change is seen as a direct challenge.  Because these new participants are not part of the traditional political elite…and because they perceive the situation in terms of threats to their centrally held values, the scope of the conflict is widened into the community at large.  Nonelite opponents of change, furthermore, are much less constrained by established political rules and norms.

⁸⁾Darcy and Laver, pp.15-16. Also, cf. Boles J., The Politics of the Equal Rights Amendment, NY: Longman, 1979, pp.17-18.

This ‘community conflict’ interpretation mirrors fairly precisely what occurred in Ireland during the 1986 referendum campaign including the “withdrawal of established elites as the dynamic element in opinion” shifted when the element of controversy heightened: once the campaign started to get ugly, government ministers began to distance themselves from it, The Irish Times reported.  Peter Barry, deputy leader of Fine Gael, foreign minister and that party’s campaign director, became “almost apologetic”, and decided to leave the country.⁹⁾cf. The Irish Times, 30 May 1986; also, cf. O’Leary and Hesketh.
Other well-known political figures such as Barry’s Fine Gael colleagues, Garret FitzGerald and Alan Dukes and the Labour Party leader, Dick Spring, had difficulty in getting airwave exposure on either television or radio, ostensibly because of the lack of balance such appearances would entail for the broadcast media – as betokened by Fianna Fáil’s ‘neutral’ stance: “Television and newspaper debates were conspicuous by the absence of established political figures, even those still active in the campaign.  As soon as most politicians ducked for cover, those who did not were marginalised.”¹⁰⁾Darcy and Laver, p.17.

Part 2: The 1995 referendum ¹¹⁾Gallagher, p.93: “In 1995, the people voted on whether to replace the absolute constitutional ban on divorce by a clause allowing divorce in cases where a couple had lived apart for four of the previous five years and there existed no reasonable prospect of reconciliation.”

The conclusion from the 1986 divorce referendum was that opponents of divorce appeared to be highly successful in playing on voters’ fears – in particular women, who as ‘deserted’ wives and mothers were pinpointed as the potential victims of such legislation.  As for the campaign itself Darcy and Laver suggested there was a “spiral of silence”, not among voters, but rather among political elites.¹²⁾Darcy and Laver, p.18
There were some indications during the 1995 divorce referendum campaign that this spiral of silence continued to affect the political elites, albeit presenting in a somewhat different guise: body language in the television appearances of prominent politicians from all political parties, tended to portray their general response to the campaign as lacklustre so far as the viewing public was concerned – in stark contrast to their official commitment to the legalisation of divorce in Ireland.   This state of affairs was seen to continue up to just a few weeks before polling day on 24 November when leaders in the main parties reacted strongly to adverse indications in opinion polls that seemed to favour a No vote, stimulating party apparatchiks into making every effort to ensure that the proposal to legalise divorce was carried in the referendum.  There was every reason for the government to be concerned.  The drop in public support was in sharp contrast to the findings of opinion polls earlier in the year which showed that support for divorce was running as high as 69%; just as in the previous divorce referendum campaign, the anti-divorce lobby was seen to be fighting vigorously against legalisation.  Government zeal in the last week of the campaign was further constrained by a Supreme Court decision in favour of Green Party MEP, Patricia McKenna, which held that the spending of state funds to unilaterally promote a Yes vote in the referendum was unconstitutional.
In the run-up to the 1995 referendum, nonetheless, there were certain developments that seemed to augur well for pro-divorce lobby.  A legislative framework had come into being, Gallagher notes, which addressed many of the issues that had led to the defeat of the proposal to legalise divorce in 1986; and a series of scandals in the Catholic church appeared to have weakened the moral authority of its continued opposition to liberalisation. In addition, this time the call for a change in the legislation was seen to be supported by all parties, rather than being simply a government initiative.¹³⁾cf. Gallagher, p.93: “…the issue did not, as in 1986, divide government and opposition; not only did the Fine Gael-Labour-Democratic Left coalition government campaign unitedly for the measure, but the main opposition party, FiannaFáil, which had been ‘neutral against divorce’ in 1986, this time officially supported the measure, though many members (including some TDs) made no secret of their intention to vote No.  The only reservations were expressed by the other opposition party, the Progressive Democrats, which wanted a complete liberalisation of restrictions on divorce and remarriage, but reluctantly backed the proposal on 24 November as at least an improvement on the existing situation.”
In the event. the electorate voted Yes for divorce, but only by the slimmest of margins, prompting the anti-divorce side to mount a legal challenge against the outcome in the High Court on the grounds that the unconstitutional spending of public money by the government could have been the decisive factor. The challenge was unsuccessful.
We now turn to the substantive part of the essay to determine the precise factors that swung the vote – albeit by the narrowest of margins – in favour of legalising divorce.  The raw data under review is based on the final pre-referendum survey of 1,000 respondents carried out by the MRBI just a week prior to polling, and analysed by the computer programme, SPSS.
Our statistical model takes the Yes vote (REFVOTE) as its dependent variable – and uses logistic regression to test hypotheses based on a number of selected independent variables that we submit would best predict the referendum outcome.

The Data
As well as being asked in the questionnaire to indicate if they would vote Yes or No in the forthcoming referendum, people were requested to say whether they were satisfied with the government and its leader, John Bruton – and if they thought that the leader of Fianna Fáil, Bertie Ahern, was doing a good job.  In addition, they were asked to state their party preferences in the event of an immediate general election.  The survey also provided data on the age, sex and the socio-economic status of respondents.
What we are attempting to establish from the final MRBI opinion poll on divorce is the best explanation of voting patterns that led to the Yes vote being carried; in simple terms – how well we can predict how people voted, with the data at our disposal.
The logit form of probability employed in logistic regression is an ideal technique for analysing dichotomous dependent variables, according to Menard.¹⁴⁾Menard, S., Applied Logistic Regression Analysis, Sage University Papers (Quantitative Applications in the Social Sciences, No. 106), London and New Delhi: Sage Publications, 1995, p.13.  The reason for using this statistical technique may be explained further, as follows: in the behavioural sciences, the use of multiple regression in multivariate analysis frequently gives rise to problems since many of the variables used to denote particular attitudes and preferences are qualitative – voting, for example in the case of political science.  To obviate such difficulties what is needed is a set of “statistical techniques that can do the work of multivariate regression but that are not subject to its liabilities in the presence of qualitative dependent variables”.¹⁵⁾Aldrich, J. and Forrest, N., Linear Probability, Logit and Probit Models, Sage University Papers (Quantitative Applications in the Social Sciences, No. 45), London and New Delhi: Sage Publications, 1984, p.10. Hence the need for logistic regression.
Again, multiple regression in multivariate analysis calls for an interval level dependent variable.  Here we clearly don’t have that – we have a categorical dependent random variable, REFVOTE (Y), that is assumed to be binary – taking on two values, 0 and 1.  For a dichotomous dependent variable such as REFVOTE, the numerical value is arbitrary and is not deemed to be “intrinsically interesting.”¹⁶⁾Menard, p.12.   For our purposes, what is intrinsically interesting is how well the classification of cases into one or other of the dependent variable, REFVOTE’s two categories (Yes/No), may be predicted by an independent variable(s). Thus, the question of interest depends on the value of the parameter, P, the probability that Y equals 1 – and may be expressed in terms of the equation,
P = P(Y = 1).¹⁷⁾Aldrich and Forrest, p.48.
In logistic regression, instead of ordinary least squares, OLS, being employed to estimate parameters – as in multiple regression – maximum likelihood estimates, MLE, are used to “maximise the value of a function” – the log likelihood function.   This tells us what the probability is of predicting observed values of Y (REFVOTE) given the values of the independent variables and parameters.¹⁸⁾Menard, p.13.   MLE may be described, therefore, as an iterative trial and error computational technique (using SPSS, in this case) that maximises the likelihood of our getting the data that we have observed.
That, in effect, is the statistical theoretical background to our logistic regression model.  As Menard says: “The solution is found by beginning with a tentative solution, revising it slightly to see if it can be improved, and repeating the process until the change in the likelihood function from one step of the process to another is negligible.”¹⁹⁾ibid. At that point the solution is said to converge.

Some further intuitive theoretical work is required, however, before embarking on the ‘tentative solution’, in deference to predicted and observed conditional probabilities: in effect, this means that some degree of reflection is advisable before commencing the iterative process – in deciding whether, for instance, to opt for a ‘bottom-up’ or ‘top-down’ model.  There are two ways to approach the problem.  We may decide to take a ‘block by block’ carefully constructed ‘bottom-up’ approach – meticulously deciding which independent variable to test against the dependent variable, REFVOTE.  Or we can opt to take as many independent variables into account as we want ab initio – discarding (or retaining) those that fail to optimise our objective as we go along.
From the viewpoint of empirical ‘integrity’, the ‘bottom-up’ approach would appear to be much the better course.  This is the course we have followed in framing the hypotheses employed below in our logistic regression – in devising a theoretical model of the voting behaviour we wish to test.

Theoretically, the order in which we rank the various hypotheses should be grounded in common sense.  In essence, it is a question of ranking the various hypotheses in order of importance – our first choice is a demographic hypothesis, AGE, as a plausible independent variable: conventional wisdom suggests that older people in liberal democracies are relatively conservative on moral/sexual matters – and divorce is a good example.  Our hypothesis therefore is that older citizens are more inclined to vote against divorce – and younger people to vote for it.²⁰⁾cf. Sinnott, R., Irish Voters Decide: voting behaviour in elections and referendums since 1918, Manchester and NY: Manchester University Press, 1995, pp.231-239. Also, cf. Gallagher, p.97.
Then common sense – in the light of the importance of ‘gender’ in the previous (1986) divorce referendum – would suggest that ‘women’ may have contributed in some substantial way to the pattern of voting behaviour that determined the referendum outcome in 1995.  Our hypothesis here is that women would tend to vote against divorce although Gallagher asserts that the ‘gender gap’, while still visible in the 1995 referendum, was “narrower than in earlier votes”.²¹⁾cf. Gallagher p.98.  He claims that in 1995 women were 5% less likely than men to favour legalisation of divorce.
The third and fourth hypotheses focus on party identification – to examine the effect that the two large conservative parties, Fianna Fáil and Fine Gael, could have on REFVOTE.  Both parties have traditionally been synonymous in general terms with the ‘illiberal’ wing of Irish society and we would therefore hypothesise that FF and FG supporters would tend to vote against divorce.  Gallagher, indeed, claims that “referendum votes have usually correlated significantly with party support”.²²⁾“In each case the FiannaFáil vote at the previous general election has correlated markedly with the conservative position (for example, r = 0.65 for the anti-divorce vote in 1986 and 0.74 for the anti-divorce vote of 1995.” p.98.
Next there is the ‘political alienation’ hypothesis to be examined, using ‘satisfaction with government’ (GOVSAT) as the independent variable.  The average citizen tends to be disgruntled with sitting governments and often is less likely to be swayed by state-backed initiatives along the lines demonstrated in the divorce referendum campaign.  Our hypothesis, therefore is that this position, too, would influence the outcome of the divorce referendum vote.  Finally, we wrap up our analysis with the urban/rural hypothesis.  A key indicator in the various public opinion surveys carried out over the last decade has been the extent of urban support for a change in divorce legislation.  It would therefore seem reasonable to include another socio-demographic variable, URBAN, too, as a predictor of the Yes vote.²³⁾cf. Gallagher p.97 and Sinnott, pp.231-239.
These independent variables go some considerable way towards explaining the voting patterns in the divorce referendum.  One could, of course, opt for a different set of variables to explain the outcome of the Yes vote.  For reasons outlined above, however, I contend that the six hypotheses chosen comprehensively explain how people voted. We now proceed to test these hypotheses.

The results ²⁴⁾It is necessary at the outset to eliminate the ‘Don’t know’ factor in the questionnaire from dependent variable, REFVOTE (how people voted), so as to reduce it to two dichotomous categories, Yes and No. Thus, when the ‘Don’t Knows’ are eliminated, the number of cases is reduced to 860 – from an original survey of 1,000 respondents. That leaves us with the values, Yes and No for the dependent variable, coded O and 1 respectively, meaning that the dependent variable is 1 (or high) if people voted No; and 0 (or low) if they voted Yes for divorce.
The log likelihood shown in the SPSS output data is “the criterion for selecting parameters in the logistic regression model”.²⁵⁾cf. Menard, pp..19-20 for details of ‘Goodness of Fit’ and ‘Log Likelihood’ data provided by SPSS: “The value of -2LL for the logistic regression model with only the intercept (constant) included is called the ‘INITIAL LOG LIKELIHOOD FUNCTION’
‘-2 LOG LIKELIHOOD’ in SPSS…The value of -2LL for the logistic regression model that includes the independent variables as well as the intercept is designated as
‘-2 LOG LIKELIHOOD’ in the ‘CHI-SQUARE’ column in the output for SPSS LOGISTIC REGRESSION.” The -2LL statistic is analogous to the ‘error sum of squares’ (SSE) in linear regression and is used in logistic regression as an indicator of how poorly the model fits “with all the independent variables in the equation”. (p.20) For explanation of Wald statistics (measure of the relationship between B coefficient and standard error) cf pp.39-40.  For explanatory details on ‘Exp (B)’ (Odds Ratio statistics), cf. p.49.  In effect what is shown by SPSS is the log likelihood multiplied by -2 which achieves approximately a chi-square distribution.   What we’re particularly interested in, however, from the point of view of our analysis, are two precise features of the SPSS partial output statistics – the ‘2 x 2’ Classification Table for REFVOTE which tells us how well we are predicting and the ‘Variables in the Equation’ table that shows how significant the impact of the independent variable equation is in terms of the dependent variable.  At the outset, we need to know whether the coefficient of the independent variable has the right sign (+ /-) so that the impact of the independent variable on voting patterns may be accurately interpreted as being for or against divorce; and also, whether or not this is statistically significant. In other words, we are testing our outlined hypotheses with reference to these particular elements.

Hypothesis 1 shows that with an overall prediction score of 61.05%, our logistic regression has achieved a relatively high prediction score for the Yes vote in the divorce referendum – but little better than ‘chance’ in predicting the No vote.  In addition, it is obvious from the ‘Variables in the Equation’ table that the B coefficient for AGE, at .2915 is highly significant (<.05); and crucially, it possesses the right sign, vindicating our hypothesis that older people tend to vote No.  This result may be viewed as the first ‘concrete block’ in our hypothetical model.

Now it’s time to look at SEX (gender), to test whether in fact the women’s vote did make a difference this time around.  Hypothesis 2, in the event, reveals an almost identical result to the first test, with the overall prediction still at 61.05% for the Yes vote and the same ‘ less than chance’ 44.82% prediction for the No vote.  It is obvious from these results that while the AGE factor remains highly significant, the B coefficient for SEX is not significant.  Therefore testing for gender has zero effect.
We are now faced with the choice of whether to leave SEX in or to remove it altogether from the equation.  On balance, we decide to leave it in.
It is not unreasonable to assume that attitudes within the two relatively large conservative Irish political parties, Fianna Fáil and Fine Gael, must reflect what is happening at the polls.
Hypothesis 3 tests the influence of the smaller of the two ‘conservative’ parties, Fine Gael, on the Yes vote in the referendum.  The result is a reduction in the overall prediction figure, to 59.53% – with a slightly ‘better than chance’ prediction of the No vote.  The sign of the B coefficient reads ‘minus’ but crucially, the results show that the Fine Gael contribution is not significant.
In Hypothesis 4, however, which measures the impact of Fianna Fáil, it is clear that this time the effect of the independent variable is highly significant.  The overall prediction score remains at 59.53%, but the prediction of the Yes vote is now almost 67%.  We may conclude from these two tests that the ‘party system’ theoretical model is working, at least as far as Fianna Fáil is concerned.  Our hypothesis has been proven to be correct.  FF supporters do in fact tend to vote No – and to a statistically significant degree.
In Hypothesis 5 we look at the broad picture of ‘political alienation’ expressed as satisfaction/dissatisfaction with government (GOVSAT).²⁶⁾Some further fine-tuning (recoding) of SPSS is required at this stage to eliminate the “no opinion’ element of the GOVSAT variable and ensure that a straightforward Y/N response to the question is reflected in the prediction statistics.  Once again we have a highly significant coefficient (.0003) – and an improved overall prediction of 61.25%.  The ‘2 x 2’ table shows our likelihood of being able to predict the Yes vote correctly to be much better than chance – and the chances of correctly predicting the No vote to be slightly better.  Once again, our hypothesis has been vindicated.
For Hypothesis 6 we attempt to gauge how accurately Urban/Rural votes predict the referendum outcome.  Again, there is a higher overall improvement in our ability to predict – to 62.53%.  The predicted Yes vote score is less than two percentage points below 70% and the predicted No vote score has also improved.   Two further points are noteworthy: first, the result is highly significant (.0000) and there is a negative coefficient sign.  This shows that people living in urban areas are more likely to vote Yes.

To summarise, we have tested our hypotheses and found that some worked remarkably well in predicting the Yes vote in the divorce referendum – the AGE, FF, GOVSAT and URBAN variables.  The other two, SEX and FG, did not work so well in that they did not tell us very much about the referendum vote.  The four variables that did predict well and which were found to significantly influence the vote go a long way to explain what happened in the 1995 referendum on divorce.

Conclusion

The 1986 referendum, Gallagher suggests, might be seen as evidence that in Ireland the referendum may sometimes be a force for conservatism.  There is strong support for this viewpoint on the basis of the opinion polls – in response to the intensity of the campaign conducted by the anti-divorce side.²⁷⁾Gallagher, p.101.   The fact that the surveys again showed majority public support for the pro-legalisation position only a few months after the 1986 referendum is at least prima facie ground for believing that referendums on balance favour conservative positions rather than representative democracy.
There are other factors to be considered, however.  For Sinnott, the overriding significance of the referendum process is that “it insulates the party system from the full brunt of secular-confessional issues by providing an alternative channel for the expression of this cleavage”.²⁸⁾Sinnott, p.295.  Gallagher, on the other hand, asserts that the fact that any change to the constitution necessitates a referendum is in itself a safeguard.  It may “inhibit change, but it undoubtedly enhances the status” of the constitution, which is seen to be controlled by the people rather than the politicians.²⁹⁾Gallagher, p.102.
So far as political science is concerned, the fact that it is possible to intuitively predict with reasonable accuracy the outcome of a referendum, using logistic regression, is reassuring.

Colman Cassidy ©