Locating Dimensions of Women’s Empowerment in Family Planning in Burkina Faso

By Tayler Nelson

Women’s “empowerment,” defined by Naila Kabeer[1] as “the expansion of people’s ability to make strategic life choices in a context where this ability was previously denied to them,” has been shown[2] to be associated with greater birth spacing, lower fertility, and lower rates of unplanned pregnancy. Yet scholars disagree[3] on how to measure women’s empowerment, and meanings of empowerment can shift across geographic and cultural contexts.

IPUMS PMA’s family planning surveys include variables that can help researchers investigate dimensions of women’s empowerment in family planning. All samples include indicators of women’s knowledge about family planning methods. Many survey rounds dig deeper, collecting data that can be used by researchers and policymakers.

The Burkina Faso 2018 Round 6 survey includes a range of variables measuring family planning attitudes, beliefs, and decision-making dynamics that relate to women’s empowerment. I used a weighted polychoric factor analysis[4] to investigate women’s empowerment in family planning in Burkina Faso. Factor analysis can help researchers reduce a large number of observed variables by identifying similar response patterns among observed variables and grouping them into a smaller set of underlying variables, or factors. Through analyzing how variables are grouped and the strength and signs of coefficients within these groups, researchers can glean insight into which sets of observed variables might be best at measuring an unobserved construct such as women’s empowerment.

After reviewing the literature on women’s empowerment in family planning, I selected nineteen PMA variables to capture dimensions of women’s empowerment in family planning in Burkina Faso. These included social context variables (URBAN, WEALTHQ, EDUCATTGEN), whether the woman is a current/recent family planning user (FPCURRECUSER), whether the woman has heard of family planning on television (FPTVHR) or radio (FPRADIOHR), and belief and attitude variables that use a Likert scale to measure how much the woman agrees with a particular statement about family planning (SAFEDISCKID, SAFEDISCFP, CONFLICTFP, DAMRELFP, NEGOTIATEKIDS, BELIEFCARRYPREG, BELIEFDAUPREG, AGREESPACE, AGREELIMIT, AGREECONTR, AGREEFP, AGREEPARTFP). I also combined two variables that measure who is the primary decision-maker in deciding to use (FPDECIDER) or not to use (DECNOFPUSE) family planning into a single ordinal variable (DECIDER) that reports how much input a woman has in the couple’s family planning decisions. I excluded not in universe (NIU) and missing cases as well as females who were not married or partnered; my analytical sample contained 1,686 women. All analyses applied appropriate sampling weights.

I retained three factors[5] and analyzed which characteristics or beliefs relate to each underlying factor by examining variable groupings and factor loadings. Table 1 displays all factor loadings above 0.3 for each variable. Looking at high factor loadings can help researchers identify underlying factors. I have marked moderate and strong factor loadings[6] in the table with an asterisk(*).

Table 1: Factor Loadings (promax rotations)

VARIABLE DESCRIPTION F1 F2 F 3
URBAN Urban/rural status .94 *
WEALTHQ Wealth score quintile .87 *
EDUCATTGEN Highest level of schooling attended .66 *
FPCURRECUSER Current or recent FP user .32
DECIDER Women’s level of say in decisions about whether to use/not use FP
SAFEDISCKID Safe to discuss when to have children w/ partner .66 *
SAFEDISCFP Safe to discuss FP w/ partner  .71 *
CONFLICTFP Conflict in relationship if used FP -.50
DAMRELFP Delaying or limiting children would deteriorate relationship w/ partner -.44
NEGOTIATEKIDS Able to negotiate w/ partner when to stop having children .65 *
BELIEFCARRYPREG Thinks a woman should not get pregnant if child still on her back .60 *
BELIEFDAUPREG Thinks a woman should not get pregnant if daughter is pregnant .48
AGREESPACE Agrees with couple that uses FP to space births .45 .44
AGREELIMIT Agrees with couple that uses FP to limit births .52
AGREECONTR Agrees with man/woman that uses contraception .34
AGREEFP Agrees with couples that use FP .59 .39
AGREEPARTFP Partner agrees with couples that use FP .69 *
FPTVHR Heard about FP on TV .66 * .38
FPRADIOHR Heard about FP on the radio .33

Factor 1 includes the majority of attitude and belief variables. I looked at similarities and differences between variables and their factor loadings[7] to interpret this factor. Variables with high factor loadings all relate to good spousal communication around family planning. For instance, SAFEDISCFP, which reports how much the respondent agrees that it is safe to discuss when to have children with her partner, has the highest factor loading. Variables related to spousal conflict, such as CONFLICTFP, have weak but negative factor loadings. Together, these results indicate that Factor 1 might be a latent measure of spousal communication. Alternative interpretations are possible: for example, it might be that this factor simply represents support for family planning, and variables with high factor loadings are the most useful observed indicators of this factor.

For Factor 2, the high factor loadings for URBAN and WEALTHQ suggest that this factor indicates socioeconomic status. The variable for whether the woman has heard about family planning on TV loads moderately highly on this factor; this seems logical assuming TV ownership is a reflection of socioeconomic status.

For Factor 3, the factor loadings are relatively small except for the loading for the belief that women should not get pregnant if she is still carrying a child on her back. In line with a Burkina Faso 2014 report that found popular acceptance of family planning for spacing rather than limiting births and stigma around women who have children too closely together, this factor might indicate adherence to traditional Burkina Faso beliefs around family planning. AGREESPACE, which reports the respondent’s agreement with couples using contraception to space the birth of their children, also loads on this factor. This factor may indicate value of traditional beliefs or more explicit support for child spacing.

This factor analysis indicates several dimensions of empowerment for understanding family planning in Burkina Faso and suggests a handful of important variables (those with high factor loadings) users may be interested in including in analyses of female empowerment. It also highlights other areas to explore. For example, I was surprised that DECIDER was not strongly associated with any single factor (though iterations of this analysis that measured joint decision-making showed high factor loading alongside the other socioeconomic status variables in Factor 2). Researchers may want to use PMA data to investigate whether spousal communication, socioeconomic status, and geographically-specific traditional beliefs remain important factors in family planning variables across different countries and years, or look further into how decision-making might align with these factors.

Get the Stata code to replicate this analysis: Stata_synatax_ipums.txt

[1] Kabeer, Naila. 2001. “Reflections on the Measurement of Women’s Empowerment.” In: Discussing Women’s Empowerment: Theory and Practice. Stockholm: SIDA: Swedish International Development Cooperation Agency.

[2] Upadhyay, Ushma D. et al. 2014. “Women’s Empowerment and Fertility: A Review of the Literature,” Social Science & Medicine 115: 111-120.

[3] Prata, Ndola et al. 2017. “Women’s Empowerment and Family Planning: A Review of the Literature,” Journal of Biosocial Science 49(6): 713-743.

[4] I used Stata’s polychoric function in order to include a mix of binary and ordinal variables in my analysis.

[5] I used Eigenvalues and scree plots to determine how many factors to retain.

[6] I am defining moderate and strong factor loadings as those loadings of .60 and greater.

[7] I looked for common conceptual threads between the highest factor loadings (asking myself, ‘What latent factor might all of these variables be measuring?’). I also looked at weak and negative factor loadings to help interpret this factor.

IPUMS IHGIS: Unlocking International Population and Agricultural Census Data

By Tracy Kugler

Nearly all countries throughout the world conduct population and housing censuses at least every ten years, and most also conduct agricultural censuses or surveys regularly. These censuses collect information on demographics, education, employment, housing characteristics, migration, agricultural land ownership, agricultural workforce, livestock, crops, and more. The resulting data can be used to study a wide range of questions, from the character of demographic transitions within and across countries, to utilization of irrigation, to educational trends among women. 

Unfortunately, this wealth of data has remained largely inaccessible to researchers. The data are typically published in reports as tables summarizing population characteristics. In recent decades, many of these reports have been published as PDF documents and made available on national statistical office websites. While the reports are available, data from a PDF document cannot be easily imported into a statistical or GIS package. Furthermore, the table structures are highly heterogeneous, both across countries and even within the same report.

The International Historical Geographic Information System (IPUMS IHGIS) is designed to provide easy access to these data in a way that researchers can easily use for analysis. In the early phases, IHGIS was known internally as “Project Mako,” named after the Mako shark, which has a global range, voracious appetite, and a reputation for a broad-ranging diet. Like the shark, IHGIS (née Project Mako) will encompass the world and ingest all kinds of data tables.

IHGIS Data

The initial version of IHGIS includes 270 tables from 9 population and housing censuses and 4 agricultural censuses. We plan to release new datasets several times a year. Our next release will include tables for an additional 12 datasets and is planned for early 2021. We have acquired over 30,000 data tables from 150 population and 107 agricultural censuses from 132 countries, which we will move through the processing pipeline over the next few years.

Datasets present/planned in the first two IHGIS data releases.
Datasets present/planned in the first two IHGIS data releases.

Our data collection efforts for population data have focused primarily on countries for which microdata are not yet available in IPUMS International. The geographic detail available with microdata is often limited due to confidentiality concerns associated with individual-level data. For several countries, notably Canada, Russia, and much of northern Europe, IPUMS International is only able to release first-level (e.g., province) identifiers. Confidentiality concerns are mitigated in summary tables. IHGIS may therefore be able to provide much more geographic detail, and we will focus on acquiring such data in future collection efforts.

You can explore the current collection through the IHIGIS data finder, where you can filter by dataset, browse available tables, select the tables you are interested in, and download the data. Your extract will include consistently structured data tables in CSV format, ready for use in your analysis. You will also receive comprehensive metadata in both human- and machine-readable formats. For more information about how to use the data finder and interpret your extract, check out our User Guide.

IHGIS also provides GIS shapefiles delineating the boundaries of the geographic units described in the data tables. Each unit is identified with a unique code in both the data tables and shapefiles, allowing you to easily join them in a GIS package.

IHGIS Under the Hood

Transforming data tables from the myriad structures in which they are published to the standardized IHGIS structure is no small task. Clearly, it would be impossible without substantial software infrastructure. But it is equally infeasible to completely automate the task of interpreting the contents of any given table. Therefore, the overarching philosophy of IHGIS data processing is to have computers do what computers are good at and have humans do what humans are good at. For example, it is relatively easy for a person to determine whether row headers identify geographic units or categories of marital status or educational attainment. Developing software to make that determination would be a significant challenge. On the other hand, having humans extract state-level totals from a table by copying and pasting is tedious, time-consuming, and error-prone.

The heart of the IHGIS data processing workflow is a table markup framework. Table markup uses Excel as an interface for a lightweight process through which researchers (mostly undergraduate research assistants) indicate the location of key structural elements within each table. For each table, students extract information such as the universe, time frame, and geographic extent. They then add keyword tags indicating the location of geographic unit headers, headers describing the characteristics summarized in the table, the table title, the extent of the data, and other structural elements.

Example of markup for a relatively simple table
Example of markup for a relatively simple table

The markup serves as a guide for our software, enabling ingest into a metadata database. The database organizes all row and column headers, titles, universes, and other metadata elements and their relationships in a consistent way. The database, in turn, enables automated restructuring of the data tables to generate the consistently structured tables in IHGIS extracts. For example, many source tables include nested geographic units at two or more levels (e.g., states and counties). IHGIS pulls the appropriate rows apart to create separate files for each level, enabling easier data linkages in GIS packages.

We hope you enjoy using IHGIS, and please send us a note at ipums@umn.edu if you have any questions, comments, or suggestions.

What’s new with IPUMS USA? Updates for Industry and Occupation Variables

By Megan Schouweiler (Senior Data Analyst, IPUMS USA) and Sophia Foster (Data Analyst, IPUMS USA)

The Census Bureau drops ACS 1-year PUMS files tomorrow (October 15, 2020)! Don’t worry, the IPUMS USA team will get right to work to get you some data as soon as possible. In the meantime, let’s talk a little about what’s new with occupation and industry variables on IPUMS USA.

New OCCSOC and INDNAICS Crosswalks Available

You may be familiar with our harmonized occupation (OCC1950, OCC1990, OCC2010) and industry variables (IND1950, IND1990). These variables harmonize occupation/industry codes based on Census Bureau classification systems to a base year, making comparisons across time much easier. Researchers are also interested in using the Standard Occupational Classification (SOC) system and North American Industry Classification System (NAICS) codes that are available in the public use data; IPUMS has not created nifty harmonized variables for these codes. We hope to harmonize these codes someday– until then, we will settle for providing great documentation about how these codes have changed over time. And we’ve recently made the documentation even better!

OCCSOC reports the primary occupation based on the SOC system, and INDNAICS reports the type of establishment of the primary occupation based on the NAICS system. Both of these coding systems are periodically updated. In the past two decades, the OCCSOC codes have been updated six times and the INDNAICS codes have been updated five times, creating a challenge for those utilizing the codes to conduct research across time. Beyond navigating the changes to the coding schemes, there are separate crosswalks for each update. We recently updated each of our crosswalks to include all iterations of the underlying coding systems from 2000 onward in a single table for OCCSOC and INDNAICS, respectively. Instead of a bunch of links to crosswalks that just compare adjacent schemes, we’ve combined all years into one table.

In total, we created four crosswalks: OCC to OCCSOC; IND to INDNAICS; OCCSOC only; and INDNAICS only. These crosswalks include detailed descriptions of how OCCSOC and INDNAICS codes have changed over time from the 2000 Census to present. Examples of changes include one occupation/industry splitting into multiple new categories, multiple categories collapsing into one occupation/industry, and updates to codes and titles. Because these types of changes occur with each new iteration of the coding scheme, it can be difficult to understand how the codes relate to one another across time. We hope that these new crosswalks provide a more comprehensive mapping of the OCCSOC and INDNAICS codes over time and will aid researchers in using these variables. These crosswalks are available to view on the IPUMS USA website and for download in both Excel and CSV format. Trust us, you’ll want to download these crosswalks to make your programming a lot easier.

Occupational Standing Variables: What are they good for?

In addition to updating the OCCSOC and INDNAICS crosswalks, IPUMS USA also released the 2018 occupational standing variables for the ACS/PRCS samples. Updated variables include OCCSCORE, SEI, HWSEI, PRENT, PRESGL, EDSCOR50, EDSCOR90, ERSCOR50, ERSCOR90, NPBOSS50, and NPBOSS90. To provide an example of how these variables can be used in research, we conducted a visual analysis using two of the updated variables, ERSCOR50 and EDSCOR50, to examine how occupational standing has changed over time for occupations highlighted during the COVID-19 pandemic.

The COVID-19 pandemic has directed particular attention towards “essential” workers and their contributions to society, raising the question of whether traditional measures of occupational standing reflect the value that we are placing on these “essential” occupations. We examined the occupational standing of occupations that have received popular attention during the COVID-19 pandemic to understand how these occupations compare to one another based on education and earnings, and to see whether these rankings have changed over time from 1950 to 2010 using the decennial Census samples.

We chose these groups from a list of occupations with high exposure to COVID-19 (Lu, 2020) and then narrowed down to a core list of occupations that have been receiving recent media attention.

Table 1
A List of Occupations Included in Each Occupation Category
Occupation Category Occupations Included in Each Category: 
Waitstaff Waiters and Waitresses
Cashiers Cashiers
Beauticians Barbers; Hairdressers, Hair Stylists, and Cosmetologists; Miscellaneous Personal Appearance Workers
Nurses Registered nurses; Nurse anesthetists; Nurse practitioners and nurse midwives; Physician assistants; Medical and health service managers
Physicians and Surgeons Physicians and Surgeons
Managers and Officials Financial analysts; Food service managers; Retail worker supervisors; Producers and Directors; Chief executives and legislators; General and operations managers; Construction managers
Teachers Elementary and Middle School Teachers; Preschool and Kindergarten Teachers; Secondary School Teachers; Education administrators
Laborers Construction Laborers; Refuse and Recyclable Material Collectors; Food Cooking Machine Operators and Tenders; Helpers–Production Workers; Subway, streetcar, and other rail transportation workers
Note. This table lists the occupations that are included in each of the eight occupation categories included in the analysis.

Next, we matched the occupation titles to the Census defined occupation categories and then to the 1950-equivalent occupation titles using the IPUMS USA variable, OCC1950. To assess occupational standing based on education and earnings, we used EDSCOR50 and ERSCOR50.

Figure 1
Figure 1

EDSCOR50 is constructed by calculating the percentage of people in a given occupation who have completed one or more years of college. ERSCOR50 is constructed by converting median earnings for each occupation to standardized z-scores, and then converting the z-score to a percent to indicate the percentage of occupations that are above or below a given occupation based on median earnings.

Overall, these figures show that the educational standing of waitstaff, cashiers, beauticians and laborers has been increasing relative to other occupations but their earnings have not. Despite their “essentialness,” examining the occupational standing variables shows that we’ve been compensating these occupations less and less over time.

Our visual analysis is just one of many ways these variables can be utilized for research. We look forward to learning about all the ways our users are leveraging occupational standing measures in their work. And remember… use if for good, never for evil!