4 Assessing Data Quality: Using Ontario Historical Records to Study the
4.3 Analysis of the Extant Records Sources
4.3.2 Demographic Characteristics
The characteristics of the individuals who could not be linked to a birth record were similar to those who could not be found in any record (Appendix D, Table 4.5). As compared to those who could be linked to a birth record, there was a higher percentage of individuals among the unlinked who were indigenous and a higher percentage with possibly itinerant occupations. A higher percentage of the unlinked died in institutions and in the North than among the unlinked and a far smaller percentage of individuals’ birth records could be found if either their father’s or their mother’s name was missing from the death record. Due to the similarities to those who could not be linked overall, a separate logistic regression for those who could not be linked to a birth record was not computed.
Those who could not be found in the 1901 census were similar to those who could not be found overall and those for whom no birth record could be found (a higher percentage were indigenous, had possibly itinerant occupations, died in institutions or in the north, and were without a father’s or a mother’s name on the death record, Appendix D). As with those who could not be found at all, those who could not be found in the 1901
census were statistically significantly older than those who could be found. Those who could not be found in the 1901 census are different from those who could not be found overall and in the birth records in that a higher percentage died in urban locations and died in Toronto. Further, fewer individuals who did not have an occupation declared on their death records were found in the 1901 census than among those who did.
There is a difference between group 1: those who could not be found overall, those who could not be found in the birth records, and those who could not be found in the 1901 census and; group 2: those who could not be found in the 1911 census, and those who could not be found in all three types of records combined. In group one, sex, marital status, and death by influenza or tuberculosis were not significantly different among those who could be linked and those who could not be linked (the significant difference for deaths by the flu in the overall linked records does not remain when other factors are controlled for in the logistic regression). If any of these factors were important, it could influence the results of this research (for example, since there are anomalous findings of sex differences among those who died by the flu globally (Chapter 2), it is important that the sex differences for those who could not be linked are small). There would be a bias problem if those who died from the flu could not be found in the same amount as those who died from other causes. There are many more significant differences between the linked and the unlinked individuals in group 2. Records appear to be more systematically biased among those who could not be found in 1911 and for those who could not be found in all three of the birth record, the 1901 census, and the 1911 census. This is partially accounted for by there being fewer individuals linked to the 1911 census, and even fewer to all three.58 In both cases in group 2, each factor on the descriptive statistics tables are significant except for death from influenza, death from tuberculosis, and the presence of a declared occupation (Table 4.11, Appendix D). Since these 2 cases are similar and there were more individuals found in the 1911 census, a separate logistic regression will only be run for those who cannot be found in the 1911 census (Table 4.11).
58
Linked to the birth record, N=2,079. Linked to the 1901 census, N=2,737. Linked to the 1911 census, N=2,424. Linked to the birth record, 1901 census, and 1911 census, N=1,667.
Table 4.11 - Descriptive Statistics of Records Linked and Not Linked to the 1911 Census. 1911 Record No 1911 Record χ2 df p Demographic Features N % N % Sex M 1,242 51.2 501 56.2 6.35 1 .01* F 1,182 48.8 391 43.8 Declared Age Mode 25 30 Median 28 29 Mean 28.5 29.3 5.88 3314 <0.01* SD 3.5 3.4 Marital Status Single 877 36.9 265 32.7 4.59 1 .03* Ever- Married 1,500 63.1 545 67.3 Indigenous Y 40 1.7 42 4.7 25.29 1 <.001* N 2,384 98.4 850 95.3 Soldier Y 70 2.9 45 5.0 9.06 1 .003* N 2,354 97.1 847 95.0 Itinerant Occupation Y 827 43.5 387 62.3 51.65 1 <.001* Nb 1,074 56.5 234 37.7 Conditions of Death Flu Y 2,067 85.3 749 84.0 0.87 1 .35 N 357 14.7 143 16.0 Tuberculosis Y 140 5.8 60 6.7 1.04 1 .31 N 2,284 94.2 832 93.3 Urban Y 1,191 49.1 509 57.1 16.41 1 <.001* N 1,233 50.9 383 42.9 Toronto Y 308 12.7 179 20.1 28.20 1 <.001* N 2,116 87.3 79.9 79.9 Institution Y 561 23.1 303 34.0 39.66 1 <.001* N 1,863 76.9 589 66.1 Region North 135 5.6 161 18.1 124.9 1 <.001* Not- North 2,289 94.4 731 82.0
State of Death Record
Occupation Y 1,345 55.5 484 54.3 .40 1 .53 N 1,079 44.5 408 45.7 Father’s Name Y 2,212 91.3 609 68.3 271.16 1 <.001* N 212 8.8 283 31.8 Mother’s Name Y 1,967 81.2 487 54.6 238.93 1 <.001* N 457 18.9 405 45.4 *P<0.05
There are many statistically significant differences among those who could not be linked in the 1911 census and those who could be linked (Table 4.11). More of the unlinked were women, were older, had ever been married, had possibly itinerant occupations, while fewer were soldiers. Greater percentages of the unlinked died in urban areas, in Toronto, in institutions, and in the north. More of the linked individuals had their parents’ names declared on their death records. The results of a logistic regression of these significant factors are found in Table 4.12. As with the overall records, the models are run in a sequential manner in order to separate the demographic characteristics of the individuals, the environment of their death, and the conditions of their death records.
Table 4.12 - Logistic Regression of Records Linked to the 1911 Census, by Demographic Features, Conditions Surrounding Death, and Conditions of the Death Record.
***p<0.01 **p<0.05 *p<0.1
Examining those who could be linked to the 1911 census as compared to those who could not, Model 1 uses logistic regression to test the demographic characteristics identified in Table 4.11. Sex is not significant, but age, being married, being indigenous, and being a soldier decrease the odds of being linked. In Model 2, these factors still decrease the odds of being linked, but now death in an urban area, death in Toronto, death in an institution,
Model 1 Model 2 Model 3 Model 4 Model 5
OR OR OR OR OR Female 1.13 1.16* 1.13 1.14 0.95 Age 0.94*** 0.94*** 0.94*** 0.94*** 0.92*** Married 0.83** 0.81** 0.83* 0.83* 0.69*** Indigenous 0.33*** 0.35*** 0.38*** 0.81 0.83 Soldier 0.47*** 0.60** 0.63** 0.63** 0.60** Urban 0.76*** 0.79** 0.79** 0.94 Toronto 0.54*** 0.74** 0.75** 0.84 Institution 0.63*** 0.71*** 0.71*** 0.77** North Region 0.33*** 0.39*** 0.45*** 0.43*** Father’s Name 1.95*** 1.85*** 1.72*** Mother’s Name 1.58*** 1.08 0.89 North Region*Indigenous 0.17*** 0.12*** Mother*Father 1.53 1.85 Itinerant 0.52*** Model χ2 66.40*** 201.50*** 283.76*** 294.31*** 262.52*** Pseudo R Square .019 .057 .081 .084 .098 AIC 3,470.53 3,343.44 3,265.17 3,258.63 2,451.53 BIC 3,506.85 3,403.97 3,337.81 3,343.37 2,538.59 N 3,144 3,144 3,144 3,144 2,450
and death in the North also decrease the odds. This is the only model in which being female increases the odds of being found, but it is significant only at α=0.1. Model 3 adds in the conditions of the death record, and both having a father’s name declared and having a mother’s name declared increase the odds of being linked, such that records with missing information are harder to link. The interaction terms for Northern Ontario and indigenous status and mothers’ and fathers’ names present are added in the fourth model. The important factor in the missing data when linking to the 1911 census is whether the fathers’ name is missing, since the mother’s name is less important for records linkage, although it is useful to have the mother’s maiden name to link to birth records. When the father’s name is missing it suggests that there is further information missing, while it was more common for the mother’s name to be missing alone. As with those who could not be linked to any record at all, it was more difficult to link an indigenous individual who also lived in Northern Ontario than to link and indigenous person who lived elsewhere in the province. The final model, Model 5, adds in the effect of possibly having an itinerant occupation. Once again, this was added last because it decreases the sample size from 3,114 to 2,450. However, as with the logistic regression of whether a record could be linked at all, the final model is better, since the AIC and BIC are smaller and the pseudo R square is larger. In this model, both death in an urban location and death in Toronto are no longer significant, but having a possibly itinerant occupation is significant. All of these factors suggest that death records that were filled out incompletely, by individuals who did not know the decedent, or by someone reporting on the death of an indigenous person were systematically harder to link. Interestingly, the interaction between indigenous status and region changes direction once itinerant status is added in Model 5. However, again, this model explains only a small percent of the variation in whether or not a record could be linked to the 1911 census (9.8%). Other factors that are likely important are the size of the household (smaller households, such as those for the newly married) are harder to link because of the fewer number of possible data points. Also, at the time of the 1911 census, the individuals in this study ranged in age from 15 to 28. This is the time in their lives when they would be most likely to be living in another household as a border or a servant, to be travelling to remote locations for work, to join the military, or to be recently married. As shown in
Chapter 6, the average size of the household among the decedents was larger in 1911 than it was for the rest of the population of Ontario. This is likely partially explained by individuals in smaller households in 1911 being harder to link than individuals in larger households.
Based on the analysis of age and the systematic biases that are found between the linked and unlinked individuals for each type of record, future analyses that require exact age at death should first consult the 1901 census. The birth records are more accurate, but have more coverage problems and the individuals who could not be linked to the 1901 census and to the birth records are similar. Further archival research should be conducted to determine the ages of those who are missing systematically, such as indigenous individuals and those from the North. While acknowledging these deficits, the 89.4% success rate of linkage to at least one other record is encouraging and conclusions can be drawn about the majority of Ontarians, especially those who are non-indigenous and live in areas other than Northern Ontario.
4.4
Conclusion
This chapter addressed the research question:
1. Through a historical demographic lens, are the extant historical records in Ontario suitable for demographic analyses of past infectious disease?
Historical demographers must use the data available in order to explain previously unexplored events in history. The historical nature of these data means that follow-up interviews to account for missing information are not possible. However, there are many extant archival sources that are underutilized and that can be drawn upon to expand and augment (or create) databases. Due to time constraints, it is likely not feasible to conduct the type of records linkage research presented in this dissertation for every question of historical interest concerning exact date of birth. Yet, the records that are available are suitable for research of this kind, although it is important to accept their limitations. For completeness, it would be best to use the 1901 census to establish age. However, for
accuracy, the birth records are more appropriate. The errors in the 1911 census mean that it should not be the first source used to ascertain date of birth, but it is useful for an understanding of the socioeconomic environment of individuals’ as they approach adulthood (Chapter 6).
Other sources could be used for this type of information, but any contemporaneous secondary source will be subject to the same types of errors. For example, the death rates by age published by the Medical Officer of Health will not include the late registrations of birth or death that had not yet been received at the time of publication. Tax assessment rolls are not good sources to obtain information about those with insufficient housing. It may be possible to find these missing individuals with different types of data sources, such as obituaries and family genealogies in future research. Despite these limitations, the extant historical records in Ontario, such as those used to create the WMMIP database, are suitable for demographic analyses of past infectious disease.