Eureka moments, such as Archimedes' sudden insight, are rare in most academic careers. Within their "situated messiness", just occasionally something we read, a conversation, a chance remark or even a fleeting thought in the bath, can lead us to do something new - or something old in a different way - and the results change our career trajectories by a few degrees. Such has been my experience. I haven't thought too much about where I'm going, I've just been lucky enough to jump in where there is excitement and reap the occasional rewards.

Early in my career at the University of Canterbury in New Zealand, light teaching and administrative loads allowed me to read widely and follow up interesting leads. Kevin Cox's 1969 essay, "The voting decision in a spatial context", was seminal. It suggested that not only the type of person you are but also where you live, and who you interact with there, can influence your vote. Geography was important in democracy's workings, it seemed.

Reading that essay initiated by far the largest shift in my career orientation. Local data facilitated critical tests designed to prove Cox wrong. It seems I was a Popperian without knowing it - while working at the university where Karl Popper penned The Open Society and Its Enemies! But the results showed that Cox was right, and I was hooked: eureka!

I read more political science, and research into biased election results in New Zealand provided further evidence of geography's importance to democracy's operations through the translation of votes into seats.

É«ºÐÖ±²¥

On leave at the London School of Economics in 1973, and while commuting by train from Croydon, I sought out inspiring books. David Butler and Donald Stokes' Political Change in Britain was certainly that. As constraints on research time tightened after I returned to the UK from New Zealand in 1974, I decided to concentrate on developing a geographical perspective within electoral studies. I soon established an amazingly fruitful partnership with Peter Taylor, whose sophisticated work with Graham Gudgin in their book Seats, Votes and the Spatial Organisation of Elections has yet to receive the praise it deserves. Pete and I met often to discuss joint research, and collaborated on a book, Geography of Elections.

Pete moved into other research areas, while I followed up some of his work by building a computer model of constituency-definition in the UK - with Economic and Social Research Council support and the programming skills of Dave Rossiter. We demonstrated that different sets of constituencies that met the Boundary Commissions' criteria could result in very disproportional electoral outcomes within individual local authorities; the Commissions were, in effect, necessarily producing "unintentional gerrymanders" because of the underlying geographies of voting. The Boundary Commission for England was sitting then, and we suggested a much better set of six constituencies for Sheffield than its recommendation: our proposal was adopted.

É«ºÐÖ±²¥

This work attracted the attention of Gerry Bermingham, a Labour councillor in Sheffield, and his legal case against the Commission cited our work as expert evidence. It was the forerunner of Labour's successful strategy to influence the next Boundary Commission review, which Dave, Charles Pattie and I enjoyed researching for five years in the 1990s, producing a well-received book, The Boundary Commissions: Redrawing the UK's Map of Parliamentary Constituencies.

In the late 1970s I had encountered papers analysing campaign spending at UK general elections and I began to explore the relationship between geographies of spending and election outcomes. Initial results were encouraging; reception of my papers was not. But I stuck with it; the evidence sustained my case that the intensity of local campaigns influenced constituency results. Others were eventually convinced. (David Denver already was.) By the late 1990s - with my research community enhanced through the Political Studies Association's Elections, Public Opinion and Parties (EPOP) group - those findings had become the conventional wisdom, and Dave Cutts extended them in a superb doctoral thesis. We had demonstrated yet again that geography matters.

Underlying much of this work was a desire to demonstrate that British voting behaviour is not just a matter of class; rather, people in similar class positions but living in different places tend to support different parties. Alan Hay and I discussed how to establish this rigorously with the available data and he identified a way forward - entropy maximising (EM).

Using EM established beyond doubt not only that many people in the same class position but living in different constituencies and regions must have supported different parties during the 1980s, but also that inter-election changes had a distinct geography. The resulting book, A Nation Dividing? Economic Well-Being, Voter Response and the Changing Electoral Geography of Great Britain, reported the first research undertaken with Charles Pattie, my main collaborator for the next two decades. (Indeed, we are still going strong. A 2006 book - Putting Voters in Their Place - synthesised our findings, but new ideas continue to generate research, not least into current party funding issues.) Work with Danny Dorling extended our ideas - using data on negative equity to test the impact of locally varying economic conditions, for example - and Danny designed excellent cartograms showing the changing geography of voting in a three-party system (which sadly didn't have the impact we thought they deserved).

Charles and I used EM in a variety of contexts to establish the extent of geographical variation - in, for example, split-ticket voting at elections for the Scottish Parliament, the Welsh Assembly and the New Zealand Parliament. Whatever we studied, we showed that geography mattered. EM asks a little more of the aggregate data in the estimation of individual behaviour patterns than Gary King's famous "resolution" of the ecological inference problem, but can more readily handle a wider range of situations.

One lunchtime conversation with Alan Hay therefore stimulated a steady trail of exploration, with results stressing the importance of where when looking at who votes what. Graduate students Ed Fieldhouse and Andrew Russell extended the ideas, their results fully sustaining the argument that geography matters, and Kelvyn Jones convinced us that multi-level modelling offered a sophisticated way to establish spatial variations at a range of scales simultaneously.

É«ºÐÖ±²¥

A substantial body of literature argued that there were also small-scale voting geographies - neighbourhood effects - but lack of data precluded exploring them until a further eureka moment. Dave, Charles, Danny and I were discussing another project in my room at the University of Bristol (an institution I joined in 1995) when Danny suggested a way to combine survey and census data to test for such effects. I recalled an earlier discussion with Graham Upton and Nick Buck at the University of Essex about something similar, and the ESRC provided money to create what we called bespoke neighbourhood data. The result was by far the strongest British circumstantial evidence that how people vote is influenced by their local contexts.

There was a real eureka moment when we finally identified the expected relationships. We had been fitting a variety of models, but it was only when several of us were sitting around a computer terminal that the anticipated pattern emerged: it may not have been the equivalent of Francis Crick and James Watson's discovery of the structure of DNA, but it was exciting nevertheless. A few years later, a chance discussion between my by-then colleague Kelvyn Jones and two Bristol economists saw us extend that work at a variety of scales. We showed that if local patterns are taken into account, regional variations are only minor. Small places matter more than big ones.

É«ºÐÖ±²¥

While researching the Boundary Commissions book, Dave remembered my earlier analyses of electoral bias. He saw the method's relevance to assessing how different an election result might have been if a Commission accepted Labour's proposals for an area rather than that of the Conservatives. It worked superbly, and established how successful Labour's strategy to influence the Commission had been: not gerrymandering but certainly very clever electoral cartography.

Our analysis of the 1997 election result using this approach identified a massive pro-Labour bias - another eureka moment when the unanticipated results emerged. If the two parties had obtained equal shares of the votes, Labour would have won 82 more seats than the Conservatives. Similar findings for subsequent elections (the pro-Labour bias was 140 seats in 2001 and 111 in 2005) firmly established the importance of geography in the operation of the UK's electoral system. It is extremely important not just how many votes a party wins but also where, which we clearly linked to our continuing work on campaigning (in our book From Votes to Seats: The Operation of the UK Electoral System since 1945). The findings have been influential, not least with Roy Jenkins' report on electoral reform, and are now being technically improved via more conversations and eureka moments, this time with Galina Borisyuk, Colin Rallings and Michael Thrasher - some of them at EPOP, others in Honiton.

Two types of eureka moment have therefore characterised my major contributions - almost all of them collaborative - in electoral studies. The first involved exploring the literature and encountering the seeds of new ideas. The second, and most profitable overall, involved conversations when throwing out interesting ideas that were often marginal to the main topic. Most led nowhere, but a few stimulated something new, with exciting results. There was never a grand scheme: I just kept reading, talking, analysing and writing. Research has been a social activity, with feedback and stimulation from a range of contexts keeping me moving forwards and occasionally steering me in new directions. (Some may wish to place all these influences together in an actor-network - perhaps with me as the central mobile mutant!)

Some of the profitable talking has happened during EPOP's annual conferences. That we (me, Charles Pattie, Dave Rossiter, Kelvyn Jones, Ed Fieldhouse, Andrew Russell, Danny Dorling, Andrew Schumann, Dave Cutts, Rich Harris, Iain MacAllister, Helena Tunstall) are geographers - or, at least, worked for a time in geography departments - was of no concern. We had established our methodological and substantive bona fides; our papers were welcomed at the meetings and in the journals. We never bothered with the rhetoric about interdisciplinarity - we just got on with it.

What conclusions have I drawn about eureka moments more generally from this experience? The first is the importance of time, which is a researcher's most crucial resource. Working through an idea calls for quality time; for unbroken periods allowing concentrated effort. Without it, much will be started but never completed. Of equal importance is "academic leisure time" - time away from the everyday demands of teaching, administration and committees - to spend reading something outside the normal, to listen to a lecture or talk to colleagues. Places are crucial for this, not least those in which spontaneous conversations can occur - such as staff common rooms, which too often nowadays are the victims of cuts, or are empty because of pressure to spend all day at a desk or laboratory bench.

Alongside time, other conditions are necessary. Money is needed for the travel that enables encounters where ideas can be developed and well-funded libraries are the infrastructural sine qua non of any academic enterprise. Will all of this still be available and possible post-Lord Mandelson, or will the unplanned serendipity that underpins so much social scientific advance die from 1,000 cuts?

É«ºÐÖ±²¥