Introduction

Administrative data provide large numbers of individuals and families for selecting birth cohorts and tracking them through time. At MCHP, both the mother's identification number - an encrypted PHIN (Personal Heath Information Number) - and the family head - usually the father, defined by the Health Insurance family Registration Number (REGNO) / (REGNO_CODE) are used to specify siblings. Children born to the same mother but different fathers can be noted as half-siblings. The accuracy of this concept has been supported by several algorithm checks as applied to nine years of birth cohorts (from 1978 to 1987, except 1983) used in MCHP research, which considered missing data, the number of children designated as having the same mother and father, and complicated blended families.

Sibling data can be used to:


• identify people/children with shared household exposures;
  
• identify people/children with varying degrees of shared genetic makeup (e.g. full-siblings, half-siblings, twins); and
  
• study the relationships between genetics, environment, family structure, etc. and education / health outcomes.

Methods to Identify Siblings in MCHP Data

Siblings are defined as children born to same mother or listed under the same adult/married REGNO. The data required for this comes from the Manitoba Health Insurance Registry data. This includes using the following variables within the Registry data:


• the mother's PHIN;
  
• the Registration Number (REGNO) / (REGNO_CODE) - this allows us to infer who the father is and allows us to track changes in family structure over time.

CAUTIONS


• REGNO belongs to family head;
  
• Children born under same REGNO to underage mothers may not be siblings: and
  
• It is important to check not only REGNO at birth but mother's PHIN.

Where both parents are known, half-siblings may be identified as children with exactly one parent in common.

Additional Information


• read the Family Structure History concept;
  
• read the Twins concept; and
  
• read the Sibling File Information - (internal access only) - related to the files and variables used to identify siblings.

Findings

As part of the work with siblings and neighbourhoods in Roos et al. (2008), same-sex sibling pairs within each neighborhood were grouped as follows:


1. Individuals classified as full siblings to the extent possible using the research registry. The same mother and father were indicated on the Vital Statistics file (incorporated into the registry). (N = 25,222 of 29,276; 86%)
   
2. Individuals likely to be siblings (possibly half siblings). The same mother was specified for each sibling, but the father was not indicated for at least one sibling. In 64 percent of the 947 sibling pairs with a mother marrying between the births of two children, the marriage took place within two years of the first sibling's birth. (N = 3,832 of 29,276; 13%)
   
3. Individuals likely to be half siblings. The same mother but different fathers, or the same father but different mothers, were noted. This category includes children where a divorce was indicated between birth of the first child and that of the second. (N = 222 of 29,276; 0.8%)

These numbers used the full nine years of cohorts. Using different-sex siblings would approximately double the numbers. Approximately 9% of the families could not be paired with another family in the same neighbourhood (Roos et al., 2008)

Significance of Sibling Studies

Sibling studies help deal with omitted variables and measurement error; they can also specify "overall background effects without limiting themselves to particular background characteristics" (Solon et al., 2000) . A readable discussion of sibling/neighborhood designs is presented by Duncan et al. (2001) . An excellent use of sibling designs is found in Oreopoulos (2003 ).

Björklund, Jäntti, and Solon (2005) explain that sibling correlations are useful in measuring "the proportion of earnings variation that can be attributed to the family and community background factors that siblings have in common" (p. 145). In their analyses, they compare sibling correlations using a wide selection of sibling types. The sibling types used in the study differ in both "their genetic connectedness and the extent to which they were reared together" (p. 146). They suggest that using a wider variety of sibling types can allow for additional leverage when determining the roles that nature and nurture have on the development of a child.

In Roos et al (2013), sibling comparisons were utilized to create a quasi-experimental design that could be used to infer causation. For further information on this design, see Lahey and D'Onofrio (2010).

Limitations / Cautions

The complaint has been made that sibling research has depended on poor data samples (Hauser & Sewell, 1986) consisting of small numbers and relatively homogenous samples. More recent research involving siblings has often utilized administrative data, allowing for the identification of siblings in a number of information-rich environments. Studying an entire population of a city or province over a number of years can generate a large and heterogeneous sample necessary for accurate estimation.

Currie et al (2010) note that any birth cohort study which is focused upon sibling comparisons is at risk of ending up with an uneven number of individuals in each cohort. Although there was a somewhat even distribution of children in each cohort at the beginning of this study, children who were in the middle cohort were more likely to have a sibling in the sample and were therefore more likely to be retained.

Oreopoulous et al (2008) discuss a threat to validity in sibling analysis:

"A threat to validity in the siblings analysis comes from changes to family or environmental contexts in between births that could account for differences in socioeconomic outcomes. By looking at a subset of siblings closer in age, fewer changes in family circumstances that may affect these outcomes are likely to occur. However, length between births is potentially endogenous (families that experience a bad first birth may wait longer to have a second child) and in this case, it might be that the effects of infant healthcare muted."

Roos et al (2014) excluded twins from the analysis of siblings. This was due to the difficulty of pairing sets of twins within a single neighbourhood, which was central to their study's design.