The results presented in this release are based on a data-matching exercise undertaken by the Central Statistics Office (CSO) from the Vital Statistics Mortality data and the 2016 Census of Population. The exercise was carried out in line with the Statistics Act 1993 and the CSO Data Protocol governing data matching exercises undertaken by the CSO.
All deaths that occurred in the twelve-month period 25 April 2016 to the 24 April 2017 were selected for matching against the Census file. The analysis was confined to persons that were usually resident and present in the State on Census Day - 24 April 2016. Deaths that were registered on certificates issued by Coroners as a result of an inquest or post-mortem examination were also included.
The overall match rate achieved was 79.8%. This is in line with similar exercises carried out in other countries such as New Zealand and Canada.
Data protocol: The results presented in this release are based on a data-matching exercise undertaken by the Central Statistics Office (CSO) of data from the Vital Statistics Mortality file and the 2016 Census of Population. The exercise was carried out in line with the Statistics Act 1993 and the CSO Data Protocol governing data matching exercises undertaken by the CSO https://www.cso.ie/en/aboutus/lgdp/csodatapolicies/csodataprotocol/
Period covered: All deaths that occurred in the twelve-month period 25 April 2016 to the 24 April 2017 were selected for matching against the Census file. The analysis was confined to persons that were usually resident and present in the State on Census Day - 24 April 2016. Deaths that were registered on certificates issued by Coroners as a result of an inquest or post-mortem examination were also included.
Data matching exercise: A total of 30,741 records from the deaths file were selected as being in scope. Of these, there was a synthetic identifier called CSOPPSN identified for 30,065 deaths. Further analysis identified 51 records as being registered twice in error leaving 30,014 available for matching to the Census 2016 anonymised dataset using the CSOPPSN. It should be noted that there were 343,626 (7.3%) of census records that didn’t have a CSOPPSN included on the Census record. There were 23,941 death records successfully matched to the Census file. Therefore, there were 6,073 mortality records unmatched. There was an overall match rate of 79.8%.
Matched v Unmatched records: Given that it was not possible to match 6,073 of the 30,014 records against the census file it is of interest to know how representative the 23,941 matched records are of the total deaths under consideration. Figure 1 presents the proportion of the unmatched death records by age-group and sex of the deceased. See Figure 1.
|Age-Group||Age-group, Male||Age-group, Female|
|80 and over||19.1||19.5|
The proportion of unmatched deaths was higher among females in the 0-24 age-group while it was higher among males in the 25 to 34 age-groups
The overall match rate achieved was 79.8%. This is in line with similar exercises carried out in other countries such as New Zealand and Canada. In New Zealand, like Ireland, an 80% match rate was achieved. See links provided directly below:
In Australia, a higher match rate was achieved. However, there was a much lower match rate for certain groups.
The fact that the unmatched deaths appear not to be randomly distributed has to be borne in mind in drawing inferences for the population as a whole based on the matched data only. The main reason why a higher match rate was not achieved was the fact that a PPSN number wasn’t provided for each Census record. In addition, there were instances where an incorrect PPSN number was provided thereby precluding complete electronic matching based on this unique identifier.
Furthermore, the absence of a post code in the mortality records also hindered matching. However, even allowing for these drawbacks, factors such as
• The quality of information on the death certificate being heavily dependent on how closely related the informant was to the deceased;
• High geographical mobility of young adults; and
• Possible under-reporting on the census
would have given rise to difficulties in achieving exact matches.
With a view to compiling death rates consistent with those published in the Vital Statistics Reports unmatched deaths were distributed pro-rata according to matched deaths for each category of a classification variable separately by broad age-group and sex. The underlying assumption is that the distribution of unmatched deaths is the same as for matched deaths for each of the variables distinguished in the release.
Life tables were produced for quintiles, disability and religion. These variables were used since they are applicable to deaths at all ages whereas data on social class and education were not available at age zero and thus not included in this release.
A detailed analysis of causes of death was not included in this release since there was a skew in the unmatched deaths (in particular for ages from 20-54) for external causes of death.
As mortality rates are strongly age-dependent it is necessary to age standardise (or age adjust) them in order to avoid drawing erroneous conclusions, especially when the results from two or more population sub-groups are being compared. In the present release age-specific mortality rates were calculated for the following broad age-groups 0-19, 20-44, 45-64, 65-74, 75-84 and 85 years and over by gender. The mortality rates compiled for these age-groups were then weighted using the 2013 European Standard Population to yield overall age standardised mortality rates. Prior to the derivation of age specific mortality rates, the unmatched mortality records were apportioned out across the age-group by gender for the various classification variables such as area of deprivation, religion etc.
It was decided to use these broad age-groups in the present release as they are more synoptic than five year age-groups and reasonably meaningful from a mortality perspective. Using five year age-groups to compile the age specific mortality rates does not give rise to any significant change to the overall age standardised mortality rates compiled using the broad age-groups described above.
High standardised mortality rates can be observed for certain residual categories which contain ‘not stated’ such as religion, level of education and social class. The explanation is primarily due to the high number of persons on the census file matched to the deaths file who were enumerated in communal establishments such as nursing homes and hospitals. Census forms received from these establishments have a higher level of not stated across all variables, probably due to the form being completed by an assistant rather than the person themselves. In the case of social class this is compounded by persons who indicate a principal economic status of ‘unable to work due to permanent sickness or disability’ as they are assigned a social class of ‘other’ thus adding to the higher level in this residual category. Accordingly, care needs to be exercised in interpreting differences in quoted mortality rates.
The census of usually resident persons present in the State on Census Night (4,689,921) was used in calculating the standardised death rates presented in Table 4(a). This population was further restricted to usually resident persons aged 15 and over and present on Census Night when calculating Table 4(b). While population was restricted to those that had ceased education in respect of Table 4(c). When producing data for Table 4(d) the population was confined to those usually resident persons present at their place of usual residence on Census Night. Finally, this population was further limited to private households only for the data presented in Table 4(e).
For the 2016 census it was possible to classify deaths by small area and then using the categorisation of small areas by deprivation score developed by Trutz Haas.
It is possible to analyse deaths by the deprivation score of the small area in which the deceased person lived prior to his or her death. The results are presented by quintiles with the first quintile representing the least deprived areas and the fifth quintile representing the most deprived areas. The underlying assumption is that if a person’s usual residence is in an area assigned a particular deprivation score then that person attracts that particular score.
Life tables were compiled for males and females which distinguished the area of deprivation in which the deceased person lived, as well as their religious beliefs and disability status. As noted earlier the life expectancy data in this release is not directly comparable to the life expectancy data presented in the Irish Life Tables No.16 (2010-2012) release that was published in July 2015 or the previous Mortality Differentials Release.
In the Irish Life Tables, the same graduation method was used, but three years mortality and population data were used, compared to one year for this release. Furthermore, since no matching was necessary, there was no bias due to unmatched deaths in the Irish Life Tables. There was a smaller number of deaths used to produce the life expectancy data for the Mortality Differentials release (i.e. only those deaths that occurred within 1 year of Census day that were matched to the Census data using a common identifier).
In the previous Mortality Differentials release, a different graduation method was used for compiling the life tables. Furthermore, a different matching method was used, based on manual matching of names and addresses. As the first name and surname were not captured as part of the census processing operation in 2006, this meant that matching with the deaths file had to be undertaken clerically. The sex, date of birth, age and place of usual residence strings were used to generate likely candidates from the census file for clerical matching.
Imputation rules applied to zero values when generating Life tables
There were some zero values for deaths at certain single year of age in respect of males and females.
To negate the issue of zero values it was decided to apply the following imputation rules:
A.1 - impute by year of age - use minimum finite value for each year of age and gender across the
A.2 - If a value is missing (i.e. no death for Single Year of Age by sex) and all variable elements (e.g. the quintiles for that Single Year of Age and sex) – then the previous year value, calculated using A1, is used.
NB: We are imputing using minimum rather than average value since the zero value represents a fact that there were no deaths in the study population. Average value would be an overestimate in this case.