THE ECONOMIC CONSEQUENCES OF PARTNERSHIP DISSOLUTION 541 Data and Methodology tion varies across waves,because not all questions are asked each survey year.'t'defines the year in which Sample these separations are observed.Since we want to com- For a detailed analysis of the economic consequences of pare the situation before and after separation,data of at partnership dissolution,it is essential to use longitudinal least one partner have to be available for years t-I data.We therefore created a thorough database of part- (before)and t (after separation).s If an individual fits nership dissolutions composed of different national these criteria,it becomes part of our study and all avail- household panels.The challenge was to make these dif- able panel information about this individual from the ferent data sets comparable across countries.In many years before and after separation is included in our data respects this worked out fine,in some respects-due to set (plus data from the year of separation).Further- limitations of the data-we had to be content with less more,we restrict the analysis to individuals between 18 perfect solutions.Table Al describes the main features of and 60 years of age to avoid a commingling of income the five national household panels used for creating our changes due to separation and income changes due to database.The Belgian,German,and British panel study retirement.Our data include 403 Belgian,1,437 German, offer a broad range of different income data,household 1,144 British,111 Italian,and 353 Swedish men and information,and socio-demographic characteristics of women,each contributing between 2 and 16 observa- each interviewed person.They were easily integrated into tions from their panel biography.Table 3 shows some our cross-national data set,while the Italian and Swedish of their characteristics. data were more problematic.The Banc of Italy Survey of As expected,we found only few cases of separation in Households'Income and Wealth,due to the initiator's Italy.Because of the few observations per individual in the interest,contains detailed information on income,but Italian panel,it is difficult to look at income changes over has rather poor data with respect to household composi- time.Therefore,Italy is not included in our multivariate tion or socio-demographic characteristics.Besides that, analyses.Another limitation of the Italian data concerns the panel section of this survey,which we use for our the type of partnership:it is only possible to identify mar- study,concerns only a small subsample of individuals, ital separations,but not break-ups of consensual unions. while the main survey consists of a series of repeated For Sweden we include both,but because of the limited cross-sections.Furthermore,individuals participating in information on marital status in the Swedish data we can- the panel section do not contribute more than three not tell these two apart.For the other countries where it is observations and then drop out of the panel.It seems as if possible to differentiate between consensual unions and not much emphasis is put on keeping the panel members marriages,the separations we look at are more often mar- motivated to take part continuously.The Swedish panel, ital break-ups than dissolutions of cohabitation. possibly because of unsteady funding,shows a lack of Furthermore,Table 3 shows that Italian and Swedish comparability over different panel waves,so that in some men and women are older on average than separating waves income data or data about the situation of the individuals in Germany,Belgium,or Great Britain at the household are missing. time of separation.For Italy,this is probably caused by Our group of interest are men and women experienc- the fact that separations included in the data set consist ing partnership dissolution.Separation is defined either of marital disruptions only,and people are usually older as separation of a consensual union or as separation of a when separating from marriage than when separating married couple.'For married couples we focus on sepa- from a consensual union.Besides that,British and Belgian ration and not on divorce because previous research individuals are more highly educated than German, (Andref and Gullner,2001)has shown that separation Swedish,and especially Italian individuals.10 is connected to more economic changes than legal Taking together part-time and full-time employment, divorce,which follows separation sometimes several Swedish (separated)women have the highest,Belgian years later,when the economic situation has already and Italian women the lowest employment rates in our stabilized.Separations were identified across the data set.It is striking that Great Britain is the only country, national panel data using similar criteria.If possible,we where almost one third of the men work part-time- combined information about marital status,different before and after a separation.Women also show the life events,and questions about household composi- highest part-time employment in Great Britain.Besides tion.In some surveys,only one of these types of infor- actual employment,it is interesting to see whether men's mation is available (e.g.,the Italian data include only and women's labour supply adapts to their new situation marital status),in others (e.g.,in Sweden)the informa- after separation.Comparing employment status beforeTHE ECONOMIC CONSEQUENCES OF PARTNERSHIP DISSOLUTION 541 Data and Methodology Sample For a detailed analysis of the economic consequences of partnership dissolution, it is essential to use longitudinal data. We therefore created a thorough database of part￾nership dissolutions composed of different national household panels. The challenge was to make these dif￾ferent data sets comparable across countries. In many respects this worked out fine, in some respects—due to limitations of the data—we had to be content with less perfect solutions. Table A1 describes the main features of the five national household panels used for creating our database. The Belgian, German, and British panel study offer a broad range of different income data, household information, and socio-demographic characteristics of each interviewed person. They were easily integrated into our cross-national data set, while the Italian and Swedish data were more problematic. The Banc of Italy Survey of Households’ Income and Wealth, due to the initiator’s interest, contains detailed information on income, but has rather poor data with respect to household composi￾tion or socio-demographic characteristics. Besides that, the panel section of this survey, which we use for our study, concerns only a small subsample of individuals, while the main survey consists of a series of repeated cross-sections. Furthermore, individuals participating in the panel section do not contribute more than three observations and then drop out of the panel. It seems as if not much emphasis is put on keeping the panel members motivated to take part continuously. The Swedish panel, possibly because of unsteady funding, shows a lack of comparability over different panel waves, so that in some waves income data or data about the situation of the household are missing. Our group of interest are men and women experienc￾ing partnership dissolution. Separation is defined either as separation of a consensual union or as separation of a married couple.7 For married couples we focus on sepa￾ration and not on divorce because previous research (Andreß and Güllner, 2001) has shown that separation is connected to more economic changes than legal divorce, which follows separation sometimes several years later, when the economic situation has already stabilized. Separations were identified across the national panel data using similar criteria. If possible, we combined information about marital status, different life events, and questions about household composi￾tion. In some surveys, only one of these types of infor￾mation is available (e.g., the Italian data include only marital status), in others (e.g., in Sweden) the informa￾tion varies across waves, because not all questions are asked each survey year. ‘t’ defines the year in which these separations are observed. Since we want to com￾pare the situation before and after separation, data of at least one partner have to be available for years t – 1 (before) and t (after separation).8 If an individual fits these criteria, it becomes part of our study and all avail￾able panel information about this individual from the years before and after separation is included in our data set (plus data from the year of separation). Further￾more, we restrict the analysis to individuals between 18 and 60 years of age to avoid a commingling of income changes due to separation and income changes due to retirement. Our data include 403 Belgian, 1,437 German, 1,144 British, 111 Italian, and 353 Swedish men and women, each contributing between 2 and 16 observa￾tions from their panel biography. Table 3 shows some of their characteristics. As expected, we found only few cases of separation in Italy. Because of the few observations per individual in the Italian panel, it is difficult to look at income changes over time. Therefore, Italy is not included in our multivariate analyses.9 Another limitation of the Italian data concerns the type of partnership: it is only possible to identify mar￾ital separations, but not break-ups of consensual unions. For Sweden we include both, but because of the limited information on marital status in the Swedish data we can￾not tell these two apart. For the other countries where it is possible to differentiate between consensual unions and marriages, the separations we look at are more often mar￾ital break-ups than dissolutions of cohabitation. Furthermore, Table 3 shows that Italian and Swedish men and women are older on average than separating individuals in Germany, Belgium, or Great Britain at the time of separation. For Italy, this is probably caused by the fact that separations included in the data set consist of marital disruptions only, and people are usually older when separating from marriage than when separating from a consensual union. Besides that, British and Belgian individuals are more highly educated than German, Swedish, and especially Italian individuals.10 Taking together part-time and full-time employment, Swedish (separated) women have the highest, Belgian and Italian women the lowest employment rates in our data set. It is striking that Great Britain is the only country, where almost one third of the men work part-time— before and after a separation. Women also show the highest part-time employment in Great Britain. Besides actual employment, it is interesting to see whether men’s and women’s labour supply adapts to their new situation after separation. Comparing employment status before