Downloadable version (PDF)

MIHA is led by the Maternal, Child and Adolescent Health Division in the California Department of Public Health (CDPH) in collaboration with: CDPH Women, Infants and Children (WIC) Division and the Center for Health Equity at the University of California, San Francisco (UCSF).

The MIHA project is supported by the California Department of Public Health using federal funds from the Title V Maternal and Child Health Block Grant and the Special Supplemental Nutrition Program for Women, Infants and Children.

Included below:

• Indicator Definitions
  
• Map of MIHA Counties
  
• Map of MIHA Regions
• Data Annotation and Suppression Criteria
  
• Weighting Methods used in publications of MIHA data
  

Indicator Definitions

Indicators are based on self-reported data from the MIHA survey and refer to the most recent birth, or pregnancy for the most recent birth, unless otherwise indicated. Unless noted, the denominator for each indicator includes all women with a live birth. Any change to a survey question or indicator, compared to how it was in a prior year, is noted in the Change in Definition and/or Comparability column(s). Indicators listed here are those that have been used in MIHA data publications since 2010.

• Prior Poor Birth Outcomes
• Health Status Before Pregnancy
• Nutrition and Weight
• Intimate Partner Violence (IPV) and Depressive Symptoms
• Hardships and Support During Pregnancy
• Substance Use
• Pregnancy Intention and Family Planning
• Infant Sleep and Breastfeeding
• Health Care Utilization and Public Program Participation
• Health Insurance Coverage
• Maternal Demographics
  

Prior Poor Birth Outcomes

Health Status Before Pregnancy

Nutrition and Weight

Intimate Partner Violence (IPV) and Depressive Symptoms

Hardships and Support During Pregnancy

Substance Use

Pregnancy Intention and Family Planning

Infant Sleep and Breastfeeding

Health Care Utilization and Public Program Participation

Health Insurance Coverage

Maternal Demographics

MIHA County-Level Data Availability, 2016-2018

Maternal and Infant Health Assessment (MIHA) Survey

Top 35 Birthing Counties
Percentage of California resident women with a live birth in 2016-2018

Data Source: 2016-2017 Birth Statistical Master File and 2018 California Comprehensive Master Birth File. 

Prepared by: Maternal, Child and Adolescent Health Division, Center for Family Health, California Department of Public Health.

MIHA Regions of California, 2016-2018

Maternal and Infant Health Assessment (MIHA) Survey

Data Annotation and Suppression Criteria

For estimates with a prevalence ≤ 50%:  

Standard error ÷ estimate

For estimates with a prevalence > 50%:

Standard error ÷ (1-estimate)

Weighting Methods

Sampling weights are created in MIHA to account for the stratified design, oversampling of specific groups, non-response among the women sampled and non-coverage of women who could not be sampled because their births were not in the sampling frame. When the final MIHA sample is weighted each year, it is designed to be representative of all mothers who delivered live-born infants in California during the calendar year that the survey was conducted and who met other inclusion criteria: the mother resided in California, was at least 15 years of age and had a singleton, twin or triplet birth. Although MIHA data are weighted to the entire birthing population, minus exclusions, the survey is only administered in English and Spanish and results may not be generalizable to women who speak other languages. The population represented by MIHA is referred to as the "target" population and is defined using the annual birth file, the Birth Statistical Master File (BSMF) (1999-2017) and California Comprehensive Master Birth File (CCMBF) (2018-present), which are the final compilation of California birth data released annually by the Center for Health Statistics and Informatics.

The MIHA survey design allows for oversampling of certain groups, meaning their probability of selection was greater than the proportion of births they represented in the state. This ensures that enough respondents participate in the survey to allow for analysis. These oversamples have included American Indian/Alaska Native women (2012-2015), Black women (all years), WIC-eligible women not participating in the WIC program (2010-2012), women with a preterm birth (2016 and later), the 20 counties with the most births (2010-2012) and the 35 counties with the most births (2013 to present).

Every woman who responded to MIHA is assigned a weight, which stands for the number of mothers in California like herself that she represents. Starting in 2010, this State Weight has consisted of 4 components (see below) calculated within strata. Additional steps have been added in subsequent years to create a Final Weight and improve the ability of the sample to represent the target population. Starting in 2011, raking (see details below) was added to the weighting process to adjust the State Weights to more accurately represent the annual birth file, particularly at the county level. Starting in 2013, trimming of weights (see details below) was implemented to reduce the influence of excessively large survey weights. These methods of raking and trimming continue to be used in all MIHA publications since 2013.

Calculation of the State Weight

The components of the State Weight are as follows: 

Non-Coverage Weight

The non-coverage weight accounts for differences between the frame from which the sample was drawn and the target population to which generalizations are made. The MIHA sample is drawn from birth certificate data for births occurring from February through May of each year, which is referred to as the "sampling frame." Birth certificate data files from which the MIHA sample is drawn are provided in monthly batches by the Center for Health Statistics and Informatics. The non-coverage weight accounts for the difference between the number of births in the sampling frame (February through May) and the number in the calendar year. The non-coverage weight also accounts for changes that might be made to the birth file after the sample is taken (e.g., births may not be in the frame files for sampling if they are reported late, but these late reported births are eventually included in the annual birth file). The non-coverage weight is defined, within stratum S, as:

Number in the Target Population _S ÷ Number in the Sampling Frame _S

Inverse of Sampling Fraction

The sampling fraction is the probability of selection, or the ratio of the number of women sampled to the number of women in the sampling frame. Therefore, the inverse of the sampling fraction within stratum S is:

Number in the Sampling Frame _S ÷ Number Sampled _S

Non-Response Weight

This weight adjusts for non-response to the survey by women who were sampled. The non-response weight is calculated within stratum S as:

Number Sampled _S ÷ Number of Respondents _S

Post-stratification Weight for Non-response (Propensity Score Adjustment)

The non-response weight described above accounts for non-response on factors used to define the strata (e.g., Black race, term or preterm birth, and county/region of residence). Additional individual-level factors may also predict whether a woman is likely to respond to the MIHA survey. Therefore, another adjustment for non-response is calculated to make the MIHA survey more representative of the target population from which the sample was taken. The probability of responding (versus not responding) is calculated using a geographically stratified logistic regression model of all sampled women. Variables in the logistic regression model come from the annual birth file and include maternal race/ethnicity, US or foreign birthplace, age, education, reported principal source of delivery payment, total children born alive, month prenatal care began, WIC participation, and term or preterm birth. A predicted probability (p) of being a respondent, or propensity score, is output for every woman sampled. The score is then re-scaled, which means that p is multiplied by a constant factor for all respondents, so that the sum of p over all respondents now adds to the number of respondents.

Starting in 2014, the post-stratification weight is capped at the 99th percentile of the post-stratification weight for each year. 

Formula for State Weight

The State Weight is calculated using the four components defined above:

NON-COVERAGE * INVERSE SAMPLING FRACTION * NON-RESPONSE * POST-STRATIFICATION

Adjustments to Create the Final Weight 

Raking Survey Weights (or Iterative Proportional Fitting)

Raking is a process by which the weighted prevalence of a selected variable is aligned with the known prevalence in a target population. In MIHA, the State Weights are raked so that weighted birth certificate variable estimates reflect the annual birth file as closely as possible at the level of the respondent's sampling region (county or group of counties). Raking is conducted over a series of predetermined variables, one at a time, in an iterative process. Raking variables include maternal age, race/ethnicity, nativity, prior cesarean section (2010-2012), low birth weight, preterm birth, prior live births, delivery payer, delivery method, BMI before pregnancy (2013-2018), education (2013-2018) and WIC participation (2017-2018). The weight assigned to each woman who falls in category C of raking variable V is multiplied by a factor of:

Number in the Target Population vc ÷ Weighted Number of MIHA Respondents vc

The first adjustment is made to the State Weight calculated in the previous section. This results in a different weight value, which is adjusted using the next raking variable and the process continues for each variable. After this is done for all desired variables, the data are checked to ensure the percentages for each raking variable are as close as possible to the annual birth file within the sampling region or group. If results can be adjusted to be more similar to the annual birth file, the process starts again with the first raking variable, using the weight from the last iteration.

After the raking process is complete, the resulting weight is rescaled (e.g., multiplied by a constant factor), so that the sum of the raked weights over all respondents adds to the number of women in the annual birth file who meet MIHA's inclusion criteria in that county/region. 

Rationale for Raking

The weighting method used before raking was introduced produced weighted data that were very close to the annual birth file at the state level and for most counties/regions. However, there were some remaining discrepancies between the weighted MIHA data and the annual birth file within subgroups of women and at the county and regional levels. Raking the State Weights produces estimates that are closer to the annual birth file for subgroups, and at the county and regional level. 

Trimming Survey Weights

MIHA weights are trimmed to reduce the influence of excessively high individual weights. Trimming is the process by which survey weights are reset to a predetermined upper limit, which reduces the occurrence of uniquely high weights that may skew survey results. To trim survey weights, a standardized distance is calculated for weights within each stratum. Strata with weights with excessive (>99^th percentile) distances are identified, and weights within those strata are trimmed at the third standard deviation (99.73^rd percentile) of weights, which are constrained to a fixed range of the original State Weight. Weights are rescaled so that totals reflect population totals in the annual birth file.

After raking and trimming, differences between county-level and regional-level MIHA data and the annual birth file are small. Very few of the estimates in the largest 35 counties are greater than 3 percentage points different from the annual birth file after raking. After trimming, the median difference between the weighted/trimmed MIHA value and the annual birth file in 2017 for raking variables was 0.00%, and 89% of estimates were within 1 percentage point of the birth file value. Differences between MIHA and the target population are sometimes greater in the smaller sampling regions, such as San Benito County, than in the counties that have more births.

Comparability across Years

Updated weighting methods have been retroactively applied to datasets starting with year 2010. MIHA publications for 2010-2012 that were produced using earlier methods (i.e., without the raking or trimming steps) have not been updated using the new weights, but all analyses published after 2013 have applied the raked and trimmed weights for 2010 MIHA data onward.  The difference between the estimates using the old and new weighting methods is small. Therefore, users may compare estimates across MIHA data years and publications, in spite of the changes to the weighting methods, unless specific changes in indicators are otherwise noted.