Abstract
The purpose of this article is to explore the connection between household composition and infant mortality in the first half of the twentieth century in Taiwan. The research results claim that illegitimate infants have higher death risk than legitimate infants; that contrary to expectation female infants have mortality rates as high as boys; and that same sex sibling order effects are approximately same among boys and girls. The significance for infant mortality of the sex of household head is the other major finding of this article: female-headed families have higher rates of infant mortality.
Introduction
Family is generally acknowledged as one of the most important components of historical, sociological, and economic analysis. In the existing literature, sex preference is a major variable in discussions of household composition and infant mortality; this variable is particularly relevant in South Asia and the developing world, where female infants commonly have higher mortality rates than males.
1
Sex alone is not the only variable to be considered in studies of infant mortality rates (IMR). For instance, Muhuri and Preston link the sex composition of older siblings with the mortality of male and female children in Matlab, Bangladesh, and find that
Girls with older sisters have 5.8 times the excess risk of dying of girls without older sisters … . The male pattern does not emerge until the index child has at least two older brothers, and the coefficient is less than half of the female equivalent … . Finally, we reiterate that girls with two or more older brothers and boys with one or more older sisters have unusually low mortality.
2
Scrimshaw used the concept of family size to estimate the survival chances of infants. In essence, the author studied the effect of fertility rates on infant mortality and concluded that “high fertility may be accompanied by the acceptance or even the unconscious encouragement of high mortality.”
3
With respect to family interest, Kikuzawa analyzed the effects of long- and short-term parental interests with respect to sex preference and childhood mortality rates in early modern Japan.
4
However, her results did not provide strong evidence to support her hypotheses:
I did find some weak evidence to support the long-term male preference hypothesis but little evidence pointing to the short-term value of girls … . From a short-term perspective, I expected that the more younger siblings girls had, the more likely they were to survive, especially in late childhood. However, I instead found a deleterious effect of younger siblings, especially younger sisters … . Under the strong influence of Confucian tradition, male preference would be observed in early modern Japan, as it is in contemporary Asian countries. However, parental interests, if there were any, did not produce clear sex difference in children mortality patterns.
5
Studies on infant mortality that account for household composition have not been done for any cities in colonial Taiwan. George W. Barclay’s Colonial Development and Population in Taiwan, one of the major English language works in this field, presents a general overview of Taiwan’s population change during the Japanese period, but does not provide specific information about Taipei. 6 New research is required to bring household composition factors to bear on changes in infant mortality in Taipei. To do so, I explore the household registers for Taipei. 7
The aim of this article is to explore the connection between household composition and infant mortality in preindustrial in Taipei. Two research approaches are employed to this end. The first involves studying the endogenous and exogenous factors. For instance, by nature, female infants have higher survival rates than males. Is this well-established biological advantage significantly modified by the strong promale bias of Taipei society?
It seems that under the Taiwanese joint household formation system, 8 the cultural preference for sons has an effect on infant mortality. As in other parts of the Chinese world, households in preindustrial Taipei were normatively joint and patrilineal. Therefore, girls could not continue the family line and surname, could not inherit property, and did not live with their natal parents after marriage. As sons were needed to perform these important roles, parents were more anxious to raise sons than daughters. In records from the early years of the Japanese period, son preference is seen to result in some cases of fatal neglect or infanticide, as well as the underregistration of female infants, affecting the sex ratio at birth, 9 and indicating that parents played key role in shaping IMR. 10
The second approach is to study the economic factors implicated in infant mortality. Traditionally, Chinese males were not only breadwinners but also successors to an ancestral line, and son preference was a strong and culturally salient norm. In the context of Taipei’s rapidly changing economy and society, however, new opportunities for girls may have raised their perceived future value to their parents. Taiwan’s people stressed both making rational use of family labor and perpetuating a patriline. 11 If a girl, who could not directly continue her patriline, nevertheless could contribute significantly to her natal family’s resources, would it raise her value, and thus her survival chances? A more rapid decline in IMR for females than for males might indicate that a changing economy was influencing attitudes and reshaping the effects of the traditional son preference. The variables used to these two approaches are: (1) sex, (2) birth status, (3) dependency ratio, (4) mother’s age, (5) birth cohort, (6) same sex sibling order, (7) sex of household head, (8) social economic status, and (9) generation type. After calculating these for all infants in the selected sample, we make use of dependency ratios as a proxy for household structure.
Background of Research Location: DaDaoCheng (大稻埕) and Meng-Jia (艋舺)
Historical data are drawn from the Japanese household registers for DaDaoCheng (大稻埕) and MengJia (艋舺): small port towns that grew into Taipei City in the nineteenth century. These towns housed few, if any, Japanese nationals. Their economies were focused on small-scale production and commerce, and residences varied in size from mansions to warrens of tiny rented rooms. DaDaoCheng and MengJia are located on the central Taipei basin and surrounded by three rivers. Given their geographical situation, DaDaoCheng and MengJia became river port towns. 12 I will refer to them simply as “Taipei.”
A proverbial Taiwanese saying, “Yi Fu, er Lu, san MengJia” (First TaiNan, second LuGang, and third Taipei), speaks to the relative prosperity of Taiwan’s past economic centers. In the second half of the nineteenth century, MengJia was Taiwan’s third largest and richest city, but at the end of the nineteenth century, tea companies moved from MengJia to DaDaoCheng because local residents were unhappy with their treatment by the tea companies. 13 MengJia, therefore, declined in economic power and interest, while DaDaoCheng grew more prominent. As the tea trade flourished, Taipei’s social structure was deeply affected, with women and children no longer restricted to free labor in the households, but instead earning wages picking tea for the tea companies.
Data and Methods: IMR and Age-Specific Mortality
The database making these calculations possible was assembled by the Program for Historical Demography, Academia Sinica. Household registers contain names of household members, certain personal characteristics (e.g., footbound, unbound, or never bound; blind, etc.), relationship to the household head, and the dates of significant events (e.g., birth, adoption, marriage, divorce, death, temporary registration elsewhere, permanent entry into and exit from the administrative unit by other means). Household heads were required to report changes to a local registry office, and the police facilitated corrections via annual home visits. The registers thus provide not only a large population sample but considerable demographic detail at the level of the individual. My access to these data distinguishes this study from Barclay’s investigations of family composition effects. His infant mortality research made use of the TaiWan ChuanKuo TiaoTsa TungChi (臺灣全國調查統計, National Survey Statistics in Taiwan) to arrive at an overall view of infant mortality in Taiwan, but that data cannot be used to analyze the effect of household composition, and differences at the individual level, on infant mortality. 14
The analysis begins by stressing the risk of death incurred by infants due to sex, birth status (legitimacy or illegitimacy), same-sex sibling order, and sex of the head of their household. To assess the effects of these variables, a logistic regression model is applied. In my data set, there are records of 9,550 births in Taipei between January 1, 1906, and December 31, 1944. In order to study the connection between infant death and household composition in a specific Taipei area, I select only infants born locally, registered, and living in their households of origin until death, or their first birthday, whichever came first. Without this stipulation, we cannot analyze the relationship between infant mortality and household composition. Adoptees must be ignored in this article, but in view of the frequency of adoption, conclude that separate research on the mortality rates of adopted infants with reference to household composition would be worthwhile. Also excluded are sojourners, who were formally registered elsewhere, and unusual households, such as those with no productive members, for example, an infant heir registered as household head. 15
The Taipei data thus defined comprise 4,283 female births, 5,267 male births, 661 female deaths, and 816 male deaths. I refer to these as “the select sample data,” and the data for all Taipei infants as “the original data.” In order to test the credibility of the morality rate of infants in the select sample, Table 1 presents the similarities of the original and the select sample data. The main consideration of the plausibility of the select sample data is those adopted and sojourners infants who were excluded from.
The Difference between the Original Data and the Select Sample Data.
Note: Birth means the number of infant who were born in Taipei. Deaths means the number of infant who were born in Taipei and then also dead in Taipei. All infants adopted out were excluded from the sample data. Neonatal deaths are the number of infant deaths in first month of life; postneonatal deaths are calculated as total infant deaths less all deaths in the first month of life. Neonatal death rate = (neonatal deaths/births) × 1000 (deaths per 1000 births); postneonatal death rate = (postneonatal deaths/postneonatal) × 1000 (deaths per 1000 births). Exogenous deaths = postneonatal deaths × 0.25; endogenous deaths = first month deaths less all exogenous deaths.
On this basis, the neonatal mortality rate is checked in both data sets. Neonatal mortality is generally considered a reflection of infant mortality without health measures and medical control. 16 There are no significant differences on male neonatal mortality rates between the original data and the select sample data. For female neonatal mortality rates, the inequalities are much less in last two periods. Furthermore, the endogenous and exogenous deaths provide a detailed check of neonatal. 17 The numbers of female exogenous deaths maintain a tiny gap between these two data. This is reasonable as the birth numbers are different. Henceforth, I refer not to “the infant mortality of the select sample” but simply to “infant mortality.”
The calculation of IMR as a ratio of deaths in the first year of life per 1,000 live births weighs all infants equally, regardless of timing of death and without respect to the time at risk. The probability of a life event, however, is closely related to the length of time at risk. I therefore adopt the age-specific mortality rate to weigh the contributions to life of infants by person-years lived. 18
In the select sample, person-years include only the time from when the individual is born in the area, to his death, or to his first birthday within a natal household remaining in the study area, whichever occurs first. All person-years relating to an infant born in the study area but leaving before the first birthday, whether through moving or adoption, and all person-years relating to infants who move into, or are adopted into, the study area after birth are excluded. Similarly, the deaths of infants are considered only for those individuals born in the study area, and remaining in the natal household in the study area until death.
Table 2 shows the probability of death calculated in terms of person-years for both the original and the select sample data sets. For the original data set, in Table 2, I count all of the infant person-years under observation, including those of infants from the time they move into, or are adopted into, the study area, and any related deaths; also included were the person-years of all those who were adopted out, or moved out, only up to the date of their leaving the area under observation. Therefore, this original data, differ from the original data in Table 1, includes 6,647 females and 6,894 males, total of 13,541 infants. Within the select data sample, the probability of death in each period before 1936 is about 30 per 1,000 greater than in the original data for females, but for males the probability of death in the two samples is approximately the same.
Age-Specific Mortality Rate (ASMR) Under Age one by Sex, 1906–1944.
Note: Person-years is calculated using the exact number of years every infant was at risk. Sex ratio = males per hundred females [(males/females) × 100]). Probability means probability of death. It is a probability an infant will die before his or her first birthday. [(Number of deaths of each period divided by person-years of each period) × 1000].
Analytic Variables
Table 3 summarizes the frequency of nine characteristics that contribute to household composition. Of these, the “sibling order” and “dependency ratio” variables require some explanation. In household registers, only legitimate infants born to the same father were arranged into an order. Illegitimate children, or children of concubines, were not assigned a same-sex sibling order number due to the low-ranked marital status of their mothers. I have rearranged the same-sex sibling position of such children to give a more complete picture of the actual (as opposed to legal) composition of the household for use in the household composition analysis.
The Frequency of Characteristics in the Selected Taipei Data, 1906–1944.
Note: (%) is the percentage of total samples. Same-sex sibling order was rearranged by adding illegitimate births to the order. Dependency ratio = [(#< age 15) + (#> age 64)]/(# age 15 to 64). Dependency ratio II = [(#<age 9) + (#> age 45)]/(# age 9 to 45). (1) <0.5: the number of dependents < (the productive members) × 0.5. (2) 0.5: the number of dependents = (the productive members) × 0.5. (3) 0.6–0.9: 0.5 × (the productive members) < the dependents < (the productive members). (4) 1: the number of dependents = the productive members. (5) >1: the number of dependents > the productive members.
In Taiwan’s colonial era household registers, the relationship of each family member to the household head changes as one household head succeeds another. This makes it difficult to classify each household into one of the four structural types usually employed (i.e., singleton, nuclear, stem, expanded). Barclay takes note of this difficulty: in a household with three members—the head, his son, and a coresident unrelated by patriline or marriage—Barclay labels the coresident an “outsider.” 19 But further investigation reveals that this “outsider” is the mother of the son listed, and the wife of the household head in an unregistered marriage. In both western and indigenous folk categories, this is simply a nuclear family. Due to this complication, Barclay did not describe the details of Taiwanese family structure with reference to each individual’s legal relationship to the household head, instead using the concept of dependency. In population research, it is common to examine the family burden using the dependency ratio, defined as the ratio of the economically dependent part of the population to the productive part. 20 As I confront this dilemma, I too use the dependency ratio as a proxy for household structure.
A higher dependency ratio indicates that the family has a heavier economic burden. Barclay notes that childhood dependency, as defined in Taiwanese census of the era, does not tell the whole story due to the prevalence of unregistered child labor.
21
The records in the census data are not appropriate for calculating family dependency ratios in Taipei. In Barclay’s statistics, the low life expectancy at age 10 justifies a different cutoff age for the end of the adult productive period.
22
Field research has shown that child labor was more common in Taipei than elsewhere in the island. An example, from field notes, is the following comment by an old man, who was plainly proud of his skill.
When I was eight years old, I worked in a tea factory in DaDaoCheng. My salary per hour was more than other children and the same as an adult, because I was good at picking tea. I picked very fast and precisely.
To match the unique economic situation and life expectancy of individuals in Taipei, I adjust the dependency ratio (setting the age terminating the dependent status of children to age 9, and setting forty-five as the age terminating the independent status of adults), named Dependency Ratio II. That is:
The dependency ratio here is divided into five categories: less than 0.5, 0.5, 0.6–0.9, 1, and more than 1. At “1,” there are as many children as adults; at “0.5,” there are half as many children as adults. Special circumstances can create unusual family structures with odd dependency ratios. It is possible, for example, that an older mother has both a newborn and a child of age fifteen. The mother’s age disadvantages her infant while the much older sibling may be an advantage for the baby’s survival. The mother’s age, therefore, is considered in this study and divided into six categories: missing, under 20, 21–25, 26–30, 31–35, and more than 35. A mother’s age may be missing from the records due to illegibility, scribal error, or residence in another registration district.
Results
This article draws on person-years analysis and regression analysis. Table 4 shows the probability of death of male and female infants with reference to several variables related to household composition. This probability, a ratio of infant deaths to person-years, indicates the mortality rate for an infant in each column. In the calculation of the relative risk of death for different periods and variables, I use the average probability of all children in each period as the reference probability.
Death Rates of Infant by Sex and Selected Characteristics.
Note: To calculate the relative mortality risk for different periods, we use the average mortality of all children in each period as the reference mortality, shown in the last column of Panel 1.
Relative risk compares the risk of death in each category with the average death risk. All children are sorted into one of the two categories of “legitimate” and “illegitimate” births. I first examine the mortality rates for illegitimate children and find that illegitimate children of both sexes have higher mortality rates than do their legitimate siblings. This was a general demographic phenomenon in Taiwan during the colonial era. Social norms stigmatizing illegitimate births, or the unfavorable economic conditions of the mothers of illegitimate children, may explain why illegitimate infants had a higher risk of death than legitimate ones. 24 I also find that while there is no difference in the degree of risk facing male or female infants when all children are included, among illegitimate children, male infants have a higher mortality rate than females. It appears that the biological disadvantage of male infants makes them especially vulnerable when subjected to the adverse conditions associated with illegitimacy.
The effect of the sex of the head of household on infant mortality is just as one might assume on the basis of Taiwan’s generally androcentric society. In an environment in which men were more likely to be effective breadwinners, infants in male-headed households were at lower risk than infants in female-headed households. I speculate that in male-headed households, if the mother died or left the family in the first year of the baby’s life, male headship alone was not sufficient to lower the infant’s mortality risk. I investigate the select sample data and find sixty-four infants who lived in a male-headed household whose mothers died, or left the household, in the first year of the baby’s life. Of these, nineteen infants died before their first birthday. Of these sixty-four infants, eight had mothers who died immediately after giving birth; of these eight infants, two died on the day of their birth. Of all these households, only three mothers who left the households did not die. As Table 5 shows, these infants had much higher probability of death compared to other infants in male-headed household. Even in a male-headed household, the unfavorable mother’s care still imposed disadvantages in terms of infant mortality. Given the importance of maternal care, I investigated IMR in female-headed households.
Death Rates of Infant in Male-headed Households Without Mother’s Care.
Both male and female infants in female-headed households had elevated risks of death compared to infants in male-headed households. The mortality rates for male and female infants in male-headed households were nearly the same from 1906 to 1944, but differed in female-headed households, with the mortality rate for boys being 1.14 times that of girls. The adverse conditions associated with a female assuming the role of household head raises the mortality risk for both sexes, but more so for males. As with the heightened risks facing illegitimate male children, this appears to be a consequence of the greater natural vulnerability of male infants. Single mothers appear to have been unable to provide the extra care needed by their male infants.
I use same-sex sibling order to analyze sex-based differences in infant mortality. Among female infants, the mortality rates of secondborn daughters are higher than those of firstborn daughters. Same-sex sibling order effects are equally evident among boys. As sons, especially firstborn sons, were viewed as family treasures, attending to their needs was always a high priority; meanwhile, firstborn daughters were favored both by biology and their early utility as assistants to their mothers. Thus, firstborn sons and daughters were measurably favored. In contrast, secondborn daughters may have been perceived as surplus mouths and received poorer care; this accords with the finding that later-born daughters were more likely to be adopted out of the household. 25
The select sample data set provides evidence that runs contrary to my assumption regarding the dependency ratio. Figure 1 shows that as dependency ratio I rises, the risk of infant death remains constant. Indeed, the highest dependency ratio surprisingly yields a relatively low mortality risk. In dependency ratio II, I adjust the ages of dependent and independent household members, matching my assumption in the period 1916–1925. The results of dependency ratio II do not support the hypothesis; thus, I must revise my initial assessment of the relationship between the dependency ratio and infant mortality.

The probability of infant death by dependency ratio I and dependency ratio II. Note: Dependency ratio I = [(#<15) + (#>64)] / (#15–64). Dependency ratio II = [(#<9) + (#>45)]/(#9–45). Probability of death = infant deaths in that period/person-years in that period.
Is it possible that a high dependency ratio among households in Taipei was characteristic of better-off households, rather than poorer ones? If so, a high dependency ratio would not imply a heightened mortality risk. The Taiwanese household registers only recorded the occupations of household heads before 1935, so the statistical analysis is applied to data drawn from the years 1906–1935. I arrange households into three strata—high, low, and unknown—according to the social standing and/or economic condition of members’ occupations, and in this way investigate the relationship between dependency ratio and the family’s economic circumstances. The results show that a higher dependency ratio I was not produced by rich households, nor did a dependency ratio II. An equal number of high- and low-income households had low dependency ratios, but many more households with a high dependency ratio had a low socioeconomic status. Marta Tienda’s article argues against my assumption. A high dependency ratio does not necessarily imply a reduction in material goods within the household, and the needs of dependents could be satisfied by sharing resources with kin in other households. 26 By sharing resources, an extended family might well improve the odds that their children would survive infancy.
Does the number of adults in a household have no influence on IMR? I also examine single-parent households or those with only one productive member. Would such household compositions lead to higher rates of infant mortality? It would seem so. In the aggregate, the dependency ratio is negatively correlated with the probability of infant death; the highest probability of infant mortality is found in household where only one adult is present (see Figure 2). More often than not, a household with only one adult will be a female-headed household, but, as described above, a male single parent would also have greater than average difficulties caring for an infant. Children in a household with two adults are exposed to less risk.

The probability of death by the numbers of adult into infant’s household. Note: Adult age is between 15 and 64.
Does the generational composition of the household, where there are three or more adults resident, have an effect on infant mortality? For example, would an infant be exposed to a different mortality risk in a nonsenior generation household? Due to the principle and limitations of the household registration records, which record only the relationship between the household head and the other members of the family, I recategorize the adult members into three categories by relationship and age: senior generation, parent generation, and junior generation. In these calculations, the senior generation includes grandparents and elders twenty years older than the infant’s mother; the parent generation includes the infant’s parents, uncles/aunts, and other adults no more than twenty years older, or fifteen years younger, than the infant’s mother; the junior generation includes the infant’s siblings who were older than fifteen years and adult members of the household who were more than fifteen years younger than the infant’s mother. In colonial Taiwan, there was a high probability that girls would marry when they were nineteen to twenty years old. 27
If the mother’s age is missing, then the father’s age is used instead. If both the mother’s age and the father’s age are missing, members older than forty are taken to be members of the senior generation; persons aged twenty to forty are assigned to the parent generation; and those aged fifteen to twenty are included in the junior generation. With reference to these three generational cohorts, five distinct household compositions are established. Type 1 is composed of three generations; type 2 includes the senior and parent generations; type 3 is made up of the parent and junior generations; type 4, the senior and junior generations; type 5, only the parent generation. Table 6 accounts for 6,970 infants who lived in households with three or more adults and presents the results of infant person-years by generational composition.
The Compositions of Generations of Households Having Three or More Adults.
Note: Generation type 1: senior, parent, and junior generations; type 2: senior and parent generation; type 3: parent and junior generation; type 4: senior and junior generation; type 5: only parent generation. The total number of infant is 6,970.
Generation type 1 presents the lowest mortality risk among all types, except type 4, but only two infants are found in the type 4 category. Their parents did not appear in their households, and their relationships to the household head were registered initially as coresidents, but later changed to grandson. The biggest group, which also had the highest probability of death, was type 2 households, in which senior and parent generations were present. Type 3 households, in which the junior and parent generations were present, had a lower IMR, suggesting that the presence of the junior generation offered infants more advantages than did the senior generation.
The lack of economic opportunities and social standing outside of their families meant that it was difficult for Taiwanese women to maintain themselves as independent household heads. This difficulty was summarized in, and perhaps exaggerated by, the popular adage “qi ping fu qui, mu ping zi gui” (妻憑夫貴, 母憑子貴, i.e., “The value of a wife depends on her husband; the value of a mother depends on her son”). Normatively, women gained their social position through attachment to men, and female household heads were supposed to be rare, except in cases of a widow holding the position between the death of her husband and her son reaching the age of maturity. 28
However, the evidence suggests that the increasing value of women’s work in a rapidly changing economy encouraged families to shift their cultural emphases. In big cities, like Taipei, female labor was in greater demand than in small cities. Even before 1930, Taipei’s women had a much broader range of occupational opportunities than did women in the rest of the island. 29 The new opportunities affected not only the island’s economy but family structures as well. When we look directly at women’s work as a possible factor in infant mortality, we begin with the understanding that Taipei mothers in this period largely worked within the home. Much of their work was housework and child care, but women also earned money through productive activities that could accommodate domestic priorities.
Women sewed professionally, made spirit money, managed tiny household businesses, helped part time in nearby food stalls, and the like. Sometimes, part of the domestic burden was taken up by the eldest daughter or other family members. I explore the possible influence of occupation and/or social economic status between 1906 and 1934 in Table 7. Obviously, the mother’s income directly affects infant’s survival. Women household heads with sufficiently high incomes could provide their infants with better living conditions, and thereby improved health. Indeed, infants born into household of high socioeconomic status had lower mortality rates than did the less fortunate. Even so, in aggregate, women household heads, even those who enjoyed a high socioeconomic status, could not provide their infants with a more secure environment than male household heads.
The Infant Mortality in Female- and Male-headed Households by Social Economic Status, 1906–1934.
Cultural alternatives to patriliny, including uxorilocal marriage, were readily available. Some parents (especially those lacking sons) kept daughters in their natal households by marrying in a son-in-law instead of marrying their daughter out. Such a marriage weakened the position of the new husband/son-in-law but was otherwise a fully effective way of perpetuating the family line of the wife’s father. This arrangement often led to the daughter eventually succeeding to her father’s place as household head. In other cases, wage-earning daughters’ marriages were postponed so that they remained within the parental households. Many of these women bore illegitimate children and only married later in life. Through this and other mechanisms, in Taipei, 586 or 10 percent of all households were female headed. In the select sample date, these 1,035 infants came from 371 female-headed households. Curiously, in the female-headed household, due to the heavy burden on livelihood and less attention on mother care of woman heads, the probability of death of heads’ infants differs from infants whose mothers were not the head of household, but not significant (see Table 8). When an infant’s mother lived in a female-headed household but not the head of the household, this infant had lower probability of death. I speculate that this infant could get better mother care because the livelihood burden was on the household head, not on his or her mother.
The Infant Mortality by the Status of Female Household Heads.
Mother’s age is an important factor in infant mortality, probably largely for biological reasons. My analysis includes mother’s age for all but 184 of the infants’ mothers, for whom full data are missing. There is no pattern and clue for the low death risk of these 184 infants. The mothers I examine ranged from twelve years to fifty-seven years. Figure 3 gives evidence of the higher risk for infant born to younger mothers. The probabilities of infant death when the mother is aged under twenty are the highest in the first two periods, and rank second in the second two periods. The higher risk for these infants is explained not just by their mother’s young age but also by the greater likelihood that they are firstborn infants. Many studies document the higher risk of firstborns independently of mother’s age. Another high-risk group is mothers who gave birth at over thirty-five years of age. For infant survival, birth to mothers between twenty-six and thirty give the best outcomes, except in one period between 1916 and 1925.

The probability of infant death by mother’s age, 1906–1944. Note: Probability of death = infant deaths in that period/person-years in that period. In 1906–1915, the probability of infant death in the category of “missing” is 0. And the number of infant in this category is 184. Missing ages for mothers occur when the mother’s name is absent from the record because of illegibility, scribal error, or residence in another registration district. There were 1,854 infants whose mother’s age was ≤20; 2,719 whose mother’s age was 21–25; 2,071 whose mother’s age was 26–30; 1,523 whose mother’s age was 31–35; and 1,199 whose mother’s age was >35.
Multivariate Analysis
In the above sections, I discuss the connection between variables of household composition and infant mortality. Are all the variables discussed above significant in determining infant mortality? In order to answer this question, I apply a logistic regression in the current research. Through a logistic regression, the strength of each variable can be measured against the others. A logistic regression can access the independent influence of every variable while synchronously controlling for the other influence. 30 This is why the logistic regression is commonly used to test various models. To do so, the infants were separated into two groups, having value 0 when the infants survived after their first year of life, and value 1 when the infants who died before their first birthday. The strength of each variable will be produced by odds ratio and the level of significance. 31
In order to understand the relationships among the variables, I set up six models (see Table 9). First three models include all the select samples (9,550 infants). Model 1 deals with infant mortality in an aggregate level, including sex, period, birth status, same-sex sibling order, dependency ratio I, mother’s age, and sex of household head. Each variable has testing purpose; sex of each infant born is accessing the gender preferences, period is for understanding the Japanese health measures, birth status, and same-sex sibling order is to test the social and culture factor, and sex of household head is for measuring mother care. I set the dividing point as 1925 for the category “period,” and two reasons support this decision. The midwives system was completed in 1923 on one hand. On the other hand, Taiwanese health facility and measurement were conspicuous from around 1925 onward. 32
Estimated Parameters of Logistic Regression of Mortality on Household Composition Variables.
Note: The total selected number of the observed infants is 6,970 in model 4, 1,035 in model 5, and 6,252 in model 6. Generation type 1: senior, parent, and junior generations; type 2: senior and parent generation; type 3: parent and junior generation; type 4: senior and junior generation; and type 5: only parent generation.
***Significant at the .001 level; **Significant at the .01 level; *Significant at the .05 level.
The structure of model 2 is the same as in model 1, but dependency ratio II is used instead of dependency ratio I in order to check the difference between age adjusted and nonadjusted. Furthermore, dependency ratio and mother’ age, birth status, and sex of head of household have highly reciprocal causation. This was caused by a phenomenon that more than one-third illegitimate children (578/1,647 = 0.35) came from female-headed households. In the select sample data, there are 1,035 infants who came from female-headed households, only 10 percent (1,035/9,950 = 0.10). To these, I exclude dependency ratio and birth status from model 3 in order to check the significance of mother’s age and sex of head of household.
The numbers of sample and purposes in models 4 to 6 are different. In these three models, I keep the infant’s sex, period, and same-sex sibling order as basic variables. Model 4 accesses the effect of the generation type on infant mortality and has 6,970 infants. Model 5 estimates the significance of the status of women household head and contains 1,035 infants. Model 6 analyzes the influence of social economic status on infant mortality and includes 6,252 infants.
In model 1, the odds ratios of dying of male infants are slightly higher than the odds ratios of female infants. And the last period presents lower odds ratios than the first one. The strong effect of period reflects the decline of infant mortality over the period. The lower risk of death in the second period undoubtedly results from improvements in the economy and environmental hygiene in Taipei over time. 33 In his article “Change of morbidity patterns in colonial Taiwan,” Shi-yung Liu stressed that public health measures and social environment made great changes to mortality. 34 But the Japanese health policy did not get a full credit in Taiwan. 35
On the other hand, the improvements made in midwifery during this period may have contributed to lowered IMR. After 1895, Japanese government introduced the midwives system into Taiwan, and issued a “TaiWan ZhongDuFu ZuCanFu JiangXiSheng GuiCheng” (台灣總督府助產婦講習生規程, Midwives regulation) in 1907. However, the infant mortality did not decrease. Until 1931, Japanese government started to pay high attention to mother and infant hygiene system, in order to ensure the manpower supply for the war. After the Japanese government established a complete midwife system in Taiwan, midwives played a positive medical role in deliveries. Midwives were also an important channel for transmitting knowledge of biomedical and hygienic practices to the population, innovations that may be assumed to have contributed further to creating wholesome environments for the newborn. 36 To summarize, the significance of period in the logistic regression denotes that the economy, environmental hygiene, and public health measures decreased the infant mortality, not only one of them.
For the variable “same-sex sibling order,” I find that second children (without regard to sex) have a higher probability of death than their siblings and that this is significant. The status of infant birth also reflects a significant influence on infant mortality. The dying of illegitimate infants were 95 to 94 percent higher than legitimate infants. In addition, the variable “dependency ratio more than 1” shows significant influence on infant mortality, apparently confirming that, in our study area, family burden correlates negatively with infant mortality. As dependency ratios rise, mortality odds fall: the death risk for an infant born into a household with a dependency ratio greater than 1 is 20 percent lower than for an infant born into a household with a dependency ratio of less than 0.5. The sex of head of household and mother’s age did not manifest any significance, excluding the unexplainable missing age of mothers.
In model 2, I set same variables as in model 1, except dependency ratio I, which is replaced by dependency ratio II. The result of dependency ratio II has little difference from dependency ratio I. Dependency ratio II less than 0.5 and more the 1 both have significant effect, and dependency ratio II also shows that a high dependency ratio followed by a low infant death rate. This strongly confirms that the dependency ratio correlates negatively with infant mortality.
There is a notable situation in model 3 where the variable “dependency ratio” and “birth status” are deleted from, mother’s age twenty-one to twenty-five is set as the reference. The female household head shows a high significance and the mother’s age younger than twenty also appears to have a significant influence on infant mortality. Disregarding the “missing mother’s age” group leaves both youngest and oldest age groups of mothers with significantly higher risks for infant mortality. For mothers aged between twenty-six and thirty-five, the risk of infant death is relatively low. Comparing model 3 with models 1 and 2, the odds of female-headed and male-headed households in models 1 and 2 are very close. But these odds become large in model 3. Besides this unlikeness, the significant effect of sibling order gets lower in model 3. When the concerning of dependency ratio and birth status was excluded, the sex of the household head and mother’s age play a fateful role in infant’s first life.
In model 4, the generation type is the main observation. Because there are only two samples in generation type 4 (no parent generation) and no infant died, the odds is less than 0.001. The odds ratio of generation type 2 (senior and parent generation) is the highest and significant. The odds of dying in generation type 2 were 32 percent higher than in generation type 5 which only had parent generation. Model 5 shows 1,035 infants in female-headed households and the status of the women head did not have crucial influence on infant death. More important, the same-sex sibling order neither had any significance. And the firstborn infants had higher odds of dying than other siblings. This states a unique situation in female-headed households. Although the same-sex sibling order neither had significance in model 6, the second infant presented a tiny higher odds of dying. The odds ratio of dying in the low social economic status is higher and significant. In model 5 the only significant variable is period, and in model 6 the social economic status is the only vital variable. In other words, when accessing the regression analysis of female-headed households with infant’s sex, period, and same sibling order, the period had a significant influence on infant mortality. When accessing the regression analysis of social economic status with infant’s sex, period, and same sibling order, the social economic status had a significant influence on infant mortality.
Conclusion
The outcomes of this research give us a new picture of infant mortality in East Asian population system. The variables of household composition, sex, birth status, mother’s age, same-sex sibling order, period, generation type, as well as dependency ratio and the sex of the head of household, are related to infant mortality in Taipei city during 1906–1944. In most populations, I find a higher male IMR, usually assumed to reflect a male biological disadvantage. But in my Taipei sample, female IMR are almost the same as those of male infants, which likely reflects the comparative neglect of daughters compared to sons. I found evidence of the male biological disadvantage in the higher male than female mortality rates when infants were subjected to the adverse circumstances related to illegitimacy and female-headed household.
The problem of underreported births and infanticide is a common phenomenon in demographic system of China. Because my sample includes only infant who can be completely observed in his or her first year of life and thus excludes the many female infants who were adopted out, the sex ratio at birth of sample is higher than the normal sex ratio. Second daughters, moreover, have lower survival chances than both their older and younger sisters, perhaps reflecting family attitudes about the value of raising additional daughters. The dependency ratio, social economic status, and the sex of the head of household are used to test the effect of family economy on infant mortality.
With respect to economy and child care, I was surprised to find a higher death risk of infants in households with lower dependency ratios. A female-headed household cannot provide infants with a higher survival chance than a male-headed household because a household with a female head was always lower social economic status. In general, the low social economic status had higher death risk than high social economic status. The household composed of three generations (senior, parent, and junior) or only parent generation provides infants with higher survival chance. When making a regression analysis of mother’s age with sex, period, and same-sex sibling order, mother’s age has more influence on infant mortality. The youngest mothers (age ≤ 20) and the oldest (age > 35) have significant and high risks of infant death.
Due to the limited length of this article, I exclude infants who were adopted in and out before their first birthday. Adoptees are an important component of the story of infant mortality in Taiwanese society; the effect of adoption on infant are dealt with in another article. The role of sex of household head is also worth further studying. Here, I have given a basic outline of the influence of sex of household head on infant mortality, but further refinements promise interesting results.
Footnotes
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author received no financial support for the research, authorship, and/or publication of this article.
