Skip to content
ALL Metrics
-
Views
-
Downloads
Get PDF
Get XML
Cite
Export
Track
Research Article

Identification of latent contraceptive ideational profiles among urban women in Senegal: Transitions and implications for family planning programs

[version 1; peer review: 2 approved with reservations]
PUBLISHED 10 May 2024
Author details Author details

This article is included in the International Conference on Family Planning gateway.

Abstract

Background

Latent ideational segmentation is an important technique that can enhance family planning (FP) communication campaigns by providing insight into prototypical “profiles” of women among heterogenous populations based on shared ideational characteristics that underpin contraceptive decision-making. This can improve the development of responsive, tailored content and help programs connect with intended audiences. In Senegal, 24% of married women who want to avoid pregnancy are not using modern contraceptive methods and in 2020, the Government of Senegal fell short of reaching its goal of increasing the modern contraceptive prevalence to 45%. Social, cultural, and cognitive factors are probable deterrents to contraceptive use. The objective of this study was to identify and interpret meaningful contraceptive ideational profiles (CIPs) among urban Senegalese women and examine how and why CIP structure, interpretation, and membership changed over time.

Methods

Using longitudinal data from 4,047 urban, in-union Senegalese women of reproductive age in 2011 and 2015, we applied latent transition analysis to identify and interpret prototypical profiles of women based on their contraceptive awareness, beliefs, self-efficacy, partner FP acceptance, partner communication, and community support.

Results

We identified four longitudinal CIPs and labeled them “CIP1: Lowest efficacy and FP awareness, highest misconceptions, unsupported,” “CIP2: Low efficacy and FP awareness, rejects misconceptions, unsupported,” “CIP3: Moderate efficacy, high FP awareness, high misconceptions, moderate support,” and “CIP4: Highest efficacy and FP awareness, fewest misconceptions, most supported.” At endline, more women were in higher-order CIPs compared to baseline. Exposure to FP communication via TV, radio, religious leaders, and health workers was associated with lower odds of membership in lower-order CIPs at endline, as was exposure to messages about FP and birth spacing.

Conclusions

This study demonstrated the potential of latent CIP methodologies to enhance current social and behavior change approaches by identifying and responding to unique and complex ideational attributes.

Keywords

latent transition analysis, latent profile analysis, contraceptive ideation, family planning, social and behavior change communication

Background

Family planning (FP) is one of the most significant public health advances of the twentieth century because it has given women greater agency to realize their reproductive intentions, that is, whether to have children, how many children to have, and when. This has advanced women’s reproductive health, education, and social and economic participation globally1,2. However, many countries have struggled to increase FP use and benefit from those health, social, and economic dividends at a population level. Despite Senegal’s early (2012) commitment to the Family Planning 2020 (FP2020) global partnership and membership in the Ouagadougou Partnership, Senegal fell short of its goals of reaching a 45 percent modern contraceptive prevalence (MCP) and reducing unmet need for modern contraception to 10 percent by 20203. In 2020, the estimated MCP was 27.7 percent among married women and 19.8 percent among all women and unmet need was 22.4 percent among married women3. A version of this article was previously published as a dissertation on our university's website here.

Increasing MCP requires dual investments in improving the supply of quality contraceptive products and services and expanding demand for these products and services. Senegal has made significant supply-side improvements in the past decade, particularly through the implementation of an informed push model of contraceptive distribution and resource mobilization48. However, demand-generation activities appear to have had less success in developing a well-informed population with a culture supportive of FP. Surveys between 2011 and 2015 of urban women and men from the Measurement, Learning & Evaluation (MLE) project in Senegal found that more than half of women and approximately half of men believed that contraceptives are dangerous to women’s health and that these perceptions did not change significantly over time”9,10. In a 2020 publication, 45 percent of Senegalese women surveyed reported that their husbands were the major source of pressure to have more children and 62 percent reported that women should not use contraceptives without their husband’s permission11. These data suggest that cognitive and social factors are probable deterrents to modern contraceptive uptake and continuation in Senegal, particularly in urban areas where supply chain barriers to FP are less pronounced.

Social and behavioral change (SBC) communication messages can increase FP knowledge, influence attitudes and social norms, and ultimately create informed, voluntary demand for FP1215. Several studies have presented evidence linking health communication programs to increased modern contraceptive use in Senegal1621. Effective SBC programming relies on audience segmentation, which traditionally identifies population subgroups based on geographic, demographic, or other characteristics that influence decision-making and predict priority outcomes22. Ideational segmentation is an important technique that can enhance FP SBC campaigns by providing insight into who to reach, how to reach them, and what messages to communicate. Ideation is defined as “the perceptions and ideas that individuals hold that reflect various social, environmental, and personal influences”23,24. Several studies have identified associations between positive health ideation and healthy behavioral outcomes including modern contraceptive use14,2529. Contraceptive ideation comprises three primary domains of psychosocial variables: cognitive (attitudes, values, knowledge, subjective norms, self-image), emotional (preferences and self-efficacy), and social (interpersonal communication, social support, and social influence)26,30,31. These psychosocial variables are foundational to women’s contraceptive decision-making. In the past decade, researchers have begun to use more specialized techniques to identify prototypical subgroups in heterogeneous populations that share key characteristics related to contraceptive ideation and use14,26,27,32,33. A latent class analysis among urban women of reproductive age in Nigeria identified prototypical classes of women who shared similar patterns of cognitive, emotional, and social indicators of contraceptive ideation29. The study found a direct relationship between membership in certain classes and contraceptive use and suggested that a better understanding of contraceptive ideational profiles (CIPs) of women may improve how FP communication programs tailor messages to different audience segments. Additional research in Nigeria, Burkina Faso, and Kenya indicated that audience segmentation based on contraceptive ideation should be routinely applied to the development of health communication programming13,14,29,34,35. Ideational profiling as a method of audience segmentation takes a person-centered approach in that it seeks out prototypical patterns of key ideational indicators such as FP awareness, beliefs, efficacy, partner communication, partner support, and community support whereas standard regression approaches often examine the contributions of these indicators in isolation by controlling for covariates. Additionally, examining whether and how these patterns change over time can reveal nuances about how sub-populations are changing and evolving. These insights can inform responsive and impactful communications campaigns.

The objectives of this study were to 1) identify and describe contraceptive ideational profiles (CIPs) of urban, partnered women of reproductive age in Senegal, 2) examine how membership in those CIPs changed over time, and 3) determine whether exposures to various health communication interventions were predictive of transitions between CIPs over time. Analyzing the relative stability of ideational profiles over time and the impact of health communication on transitions between more empowered and less empowered ideational profiles can help communication programs better tailor FP message content and modalities. To our knowledge, this was the first application of latent transition analysis to advance the understanding of contraceptive ideation in a low-resource setting.

Methods

We analyzed data from the Measurement, Learning and Evaluation (MLE) project for the Urban Reproductive Health Initiative in Senegal that included survey data from a longitudinal (2011–2015) sample of urban Senegalese women of reproductive age (15–49) who lived in Dakar, Guédiawaye, Pikine, Mbao, Mbour, and Kaolack10,21. In 2011, a two-stage sampling design was used to collect data from a random sample of 21 households in 268 study clusters, stratified by poor and non-poor, based on a probability proportional to their populations. Poor strata were oversampled to increase inclusion of poor households and women. All women ages 15–49 years in the selected households were eligible to participate in the 2011 baseline survey. At baseline, 9,614 women were interviewed and 6,927 of these women were re-interviewed at endline (2015) for a follow up of 73.5%10,21. We restricted our dataset to 4,047 women who were in-union (married or living with a partner) at baseline and who responded to the endline survey. Several important survey questions were only asked of women who were in-union because premarital sex is taboo in Senegal. Additionally, initial exploratory latent profile analyses at baseline found no significant or meaningful differences when comparing profiles of women without endline data to profiles of women with endline data. A balanced dataset allowed us to clearly articulate and quantify movement between CIPs over time and endline probability weights accounted for attrition.

To estimate and characterize transitions between latent profiles over time, we conducted a latent transition analysis (LTA). LTA is a form of finite mixture modeling often used in health policy and health services research to identify unobservable, or latent, prototypical profiles within a large heterogeneous population based on observed response patterns and to assess how profiles and profile membership changes over time3639. LTA assumes that there is a mixture of discrete distributions that comprise the population heterogeneity and identifies the most likely model by defining a finite number of profiles40. LTA assumes an underlying latent categorical variable that gives rise to the observed variables and returns the probability of membership in each profile based on patterns of item-response probabilities. LTA has become an important person-centered tool for examining changes and trends in groupings of health and behavioral characteristics41,42.

We selected ideational indicators (constructs) of the latent contraceptive ideational profile (CIP) variable based on 1) Kincaid’s theoretic framework of contraceptive ideation, 2) evidence from the literature on important factors in FP decision-making, 3) previous research on contraceptive ideational profiles in low-resource settings, and 4) a review of the shape and variance of available variables in the dataset29,34,35. Wurpts and Geiser (2014) suggest that using at least five indicators with strong relationships to the latent profile variable contributes to greater certainty in defining the classes43. After considering tradeoffs between including more indicator variables and increasing the risk of data sparseness and including a subset of indicator variables that risk not adequately specifying the model, we ultimately selected six indicators for which there is substantial evidence of their importance in contraceptive decision-making.

Once we selected theory- and evidence-based domains and indicators of our latent variable, we created constructs using the available variables in the MLE dataset (Table 1). Three indicators were continuous constructs comprised of multiple survey questions. For FP awareness, women were asked to name any FP methods they knew. When they had listed all methods they knew, they were then prompted on whether they had heard of other methods that they had not spontaneously listed. For each method that a woman mentioned spontaneously, she received a score of 1 point. For each method that she had heard of when prompted, she received 0.5 points. Points were summed and the resulting construct had a range of 0 to 13. For misconceptions, women were asked whether they totally agreed, agreed, disagreed, or totally disagreed with seven statements about common negative beliefs and misconceptions about FP. Responses received -2, -2, 2, and 2 points, respectively, such that high scores indicated a strong rejection of misconceptions. Perceived self-efficacy was scored in the same way for eight questions about a woman’s beliefs about her ability to act on her contraceptive intentions. For constructs that were a composite of multiple variables, we examined Cronbach’s alpha to test whether the variables reliably measured the same phenomenon and found satisfactory internal consistency.

Table 1. Overview of selected indicators of contraceptive ideation.

Six indicators were developed including FP Awareness, Rejection of FP Myths and Misconceptions, Perceived Self-Efficacy, Partner FP acceptance, Partner Communication, and Perceived Social Support for FP.

Indicator & Rationale for InclusionIndicator DescriptionSurvey Questions
FP Awareness

Women who have low levels of knowledge or
awareness of contraceptive methods are unable
to make informed decisions around contraceptive
uptake and continuation11,44,45
All contraceptive methods mentioned
unprompted received a score of 1; methods
acknowledged after interviewer prompting
received a score of .5. Scores were summed
across contraceptives to create an overall
awareness indicator of 0 to 13. Higher
values indicate higher awareness of
methods.
Baseline
Range: [0, 13]
Mean (SE): 6.14 (0.03)
Cronbach’s alpha: 0.72
Endline
Range: [0, 13]
Mean (SE): 6.92 (0.08)
Cronbach’s alpha: 0.77
Heard of female sterilization
Heard of male sterilization
Heard of oral pills
Heard of IUD
Heard of injectables
Heard of implants
Heard of male condoms
Heard of female condoms
Heard of emergency contraception
(EC)
Heard of rhythm method
Heard of withdrawal
Heard of spermicide
Heard of lactational amenorrhea
method (LAM)
Rejection of FP Myths and Misconceptions

Women who hold misconceptions or negative
beliefs about contraception are less likely to use
FP9,4649
This indicator summarizes responses to
seven questions about common myths and
misperceptions about FP
Possible responses included: totally agree
(-2), agree (-1), don’t agree (1), and totally
disagree (2). Responses were summed to
create an indicator with values ranging from
-14 to 14. Higher values indicate stronger
rejection of myths and misconceptions.
Baseline
Range: [-14, 14]
Mean (SE): 3.03 (0.30)
Cronbach’s alpha: 0.90
Endline
Range: [-14, 14]
Mean (SE): 5.21 (0.20)
Cronbach’s alpha: 0.87
Being injected with a contraceptive
product makes a woman permanently
sterile
People who use contraceptives end
up having health problems
Contraceptives can harm the uterus
Contraceptives reduce sexual desire
Contraceptives can cause cancer
Contraceptives can cause birth
defects
Contraceptives are dangerous for
one's health
Perceived Self-Efficacy

Higher levels of confidence in one’s ability to take
action is positively associated with contraceptive
intention to use, uptake, and continued use5055
The contraceptive efficacy beliefs indicator
uses eight questions about a woman’s
ability to act in her own interest regarding
FP use. Possible responses included:
totally agree (2), agree (1), don’t agree (-1),
and totally disagree (-2). Responses were
summed to create an indicator with values
ranging from -16 to 16. Higher values
indicate higher perceived self-efficacy.
Baseline
Range: [-16, 16]
Mean (SE): 5.69 (0.22)
Cronbach’s alpha: 0.81
Endline
Range: [-16, 16]
Mean (SE): 5.06 (0.24)
Cronbach’s alpha: 0.79
Able to initiate conversation about FP
with partner
Able to convince partner that you
should use FP
Able to go to a place where FP is
available to get an FP method if
desired
Able to obtain a method
Able to use a method even if partner
doesn’t want to
Able to use a method even if none of
your friends or neighbors use one
Able to use FP even if your religious
leader did not approve
Able to use a method even if have
side effects
Partner FP acceptance

Male partners can influence a woman’s uptake
of contraception based on her perceptions of his
approval or disapproval11,5661
This indicator represents whether a partner
forbids FP use (does not forbid=1).
Baseline
Range: [0, 1]
Mean (SE): 0.81 (0.01)
Endline
Range: [0, 1]
Mean (SE): 0.83 (0.01)
Does your partner forbid FP?
Partner Communication

Women’s perception of the acceptability
and normalcy of communication about FP
between husbands and wives can influence her
contraceptive decision-making57,58,62,63
The partner communication indicator is a
binary indicator that indicates whether a
woman has discussed FP with her partner in
the last six months (yes=1).
Baseline
Range: [0, 1]
Mean (SE): 0.41 (0.01)
Endline
Range: [0, 1]
Mean (SE): 0.36 (0.01)
Have you discussed FP with your
partner in the last six months?
Perceived Social Support for FP

Women who feel comfortable discussing FP with
their spouse, friends, or family, or who believe that
there is community acceptance of FP are more
likely to use contraception46,47,6466
The Social Support indicator includes one
question about perceived support for FP
use by community members (yes=1)
Baseline
Range: [0, 1]
Mean (SE): 0.50 (0.02)
Endline
Range: [0, 1]
Mean (SE): 0.57 (0.02)
Are there people in your community
who would congratulate you for using
FP?

Baseline and endline estimates account for survey probability weights and clustering.

Figure 1 visualizes the latent transition conceptual model for contraceptive ideation. In this model, we hypothesize that there is a categorical latent variable that explains similarities in contraceptive ideation among subgroups of women. We also hypothesize that exposures to various types of FP health communication programming may differentially affect membership in the endline CIPs and transitions between baseline and endline CIPs.

dc81af7b-4ceb-49f9-9960-8b51102586d8_figure1.gif

Figure 1. Conceptual model of latent transition analysis with health communication covariates.

This model hypothesizes that there is a categorical “latent CIP” variable that explains similarities in cognitive, emotional, and social ideational indicators among subgroups of women. It also hypothesizes that exposures to various types of FP health communication programming may differentially affect membership in the endline CIPs and transitions between baseline and endline CIPs.

Model specification and selection

Using six indicators of contraceptive ideation, we estimated longitudinal latent models with two to eight CIPs. We relied on Akaike information criteria (AIC), Bayesian information criteria (BIC), and sample-size adjusted BIC (ABIC) to make relative decisions about model fit39,40,67. We considered entropy, which is a standardized index of assignment accuracy where higher values indicate a more accurate assignment of individuals to latent profiles and better separation between those profiles68. We also considered the proportional membership in each status as well as the utility of the interpretations for FP communication programs.

After selecting the optimal number of CIPs, we tested our hypothesis of measurement invariance by constraining the item-response probabilities to be equal at baseline and endline. In LTA models, there is measurement invariance across two time points when the relationship between the latent variable and the observed constructs are the same, even if the distribution of the latent variable is different across time points39. Conceptually, it is useful to have measurement invariance because it assures that latent CIPs can be interpreted the same way over time and across groups. To test for measurement invariance we conducted a likelihood ratio test, which indicates whether imposing equality constraints significantly changes model fit69. Practically, constraining item-response probabilities also stabilizes estimation and improves identification by decreasing the number of estimated parameters39.

We estimated three sets of parameters: latent CIP prevalences, item-response probabilities, and transition probabilities. Once the model was identified, we imposed parameter restrictions so that the item-response probabilities were identical in both time periods, allowing us to maintain consistent profiles across time. We calculated transition probabilities to help us understand where women in a certain CIP at baseline transitioned at endline.

Once we understood the structure and interpretation of our LTA model, we introduced covariates to determine whether exposures to FP messages were associated with endline CIP membership and transitions between different CIPs over time. At endline, women were asked whether they had heard or seen FP messages from mass media modalities in the past 3 months and community-based sources in the past 12 months. All questions about FP exposures were coded as binary variables. We also developed a count variable [0–11] that summed all reported modalities to determine whether additional modalities of FP messaging influenced transitions over time. Finally, women were also asked about the content of the FP messages that they had heard. We did not include wealth or age because those covariates were not significant predictors of baseline or endline profile membership in initial exploratory latent profile analyses. Because Models 1 and 3 included covariates representing similar constructs (e.g., heard about FP through radio, TV, news, etc.), we calculated correlation tables across all included covariates to determine whether there was significant overlap in the general population. For Models 1 and 3 the highest correlations between covariates were 0.41 and 0.31, respectively, indicating sufficiently distinct indicators.

Covariates are incorporated into LTA using a multinomial logistic regression framework that estimates the effect for each latent CIP in comparison to a reference class39. We estimated three multinomial logistic regression models to test whether FP communication modalities, multiple modalities, and message content influenced membership in endline profiles. We selected covariates in the model to which at least 10 percent of surveyed women reported exposure. In Model 1, we included exposures to FP information via TV and radio from the mass media category and religious leaders and community conversations with community health workers (CHW) from the community category. We also controlled for education, city, and parity, all of which proved to be significant in predicting profile membership in earlier exploratory modeling. In Model 2, our covariate of interest was the count variable of unique communication modalities as well as the control covariates. In Model 3, we included whether women had reported hearing messages about spacing births, the legitimacy of FP, spousal communication, the position of Islam on FP, limiting family size, or rumors and fears about FP, as well as the control covariates.

Next, we ran three multinomial logistic regression models which estimated transitions between latent CIPs over time, conditional on baseline CIP membership. To address misclassification, we used the manual BCH three-step approach for longitudinal LTA7072. The three-step approach ensures that covariates do not influence the measurement model of latent contraceptive ideation and the BCH weighting approach accounts for error in classification. LTA allows for the incorporation of complex survey design features and we applied endline six-city probability weights that account for attrition in the endline sample and adjusted for clustering.

Results

LTA model selection considerations included model fit, parsimony, interpretation, and utility for answering our research questions of interest. We ultimately selected the LTA model with four CIPs at baseline and endline because it represented improved model fit over the two- and three-CIP models producing lower log likelihood, AIC, BIC, and ABIC statistics. We did not select the five-CIP or higher models because they had issues of extreme sparseness in their contingency tables, with some CIPs having no members. Sparseness is problematic for examining how key covariates influence transitions between CIPs over time. Although we did not identify measurement invariance over time, we still found it useful to constrain the item response probabilities so that the interpretation of the CIPs was the same at baseline and endline. Measurement variance may have been due to overall improvements in contraceptive ideation over time, as indicated by overall increases in the scores of indicators at endline. The constrained model fit statistics represented only slight decreases in model fit compared to the unconstrained models but improved entropy.

Once we selected the model, we analyzed the item response probabilities for each of the six indicators to interpret each of the four latent CIPs. Table 2 presents the mean item response probabilities for each indicator for the constrained four CIP LTA model. For all item response probabilities, CIP 1 had the lowest (least desirable or least empowered) scores across all indicators, though some were not statistically different from CIP 2. CIP 4 had the highest (most desirable, most empowered) scores across all indicators, with no overlap of confidence intervals with other CIPs. CIPs 2 and 3 were qualitatively different in that women in CIP 2 were more likely to reject negative beliefs and misconceptions than women in CIP 3 but otherwise had lower FP awareness, self-efficacy, partner support, partner communication, and social support. CIP 1 had the smallest proportional membership at baseline and endline. At endline, the proportional memberships of CIP 1 decreased and CIP 4 increased compared to baseline, indicating improvements in contraceptive ideation over time.

Table 2. Constrained mean item response probabilities for selected LTA model at baseline and endline.

For all item response probabilities, CIP 1 had the lowest (least desirable or least empowered) scores across all indicators, though some were not statistically different from CIP 2. CIP 4 had the highest (most desirable, most empowered) scores across all indicators, with no overlap of confidence intervals with other CIPs.

Indicator
[Response range]
CIP 1
Mean [95% CI]
CIP 2
Mean [95% CI]
CIP 3
Mean [95% CI]
CIP 4
Mean [95% CI]
FP awareness
[0, 13]
3.94
[3.47 – 4.42]
5.39
[4.88 – 5.90]
6.54
[6.36 – 6.72]
7.10
[6.97 – 7.23]
Rejects misconceptions [-14, 14]-6.35
[-8.09 – -4.60]
5.49
[4.79 – 6.18]
-2.93
[-3.43 – -2.42]
8.34
[7.91 – 8.77]
Perceived self-efficacy [-16, 16]-8.77
[-10.24 – -7.30]
0.16
[-1.14 – 1.46]
5.76
[5.23 – 6.28]
8.09
[7.62 – 8.55]
Partner Support
[0,1]
0.48
[0.39 – 0.57]
0.62
[0.54 – 0.69]
0.80
[0.76 – 0.84]
0.93
[0.90 – 0.96]
Partner communication [0, 1]0.01
[0.00 – 0.03]
0.08
[0.02 – 0.14]
0.35
[0.31 – 0.39]
0.55
[0.51 – 0.58]
Social support
[0,1]
0.14
[0.08 – 0.20]
0.20
[0.12 – 0.27]
0.47
[0.42 – 0.52]
0.71
[0.67 – 0.75]
Proportional baseline membership0.070.120.340.47
Proportional endline membership0.040.170.190.60

Figure 2 presents a standardized z-score distribution of the constrained latent CIP indicators by CIP using baseline population means and standard deviations. This figure visualizes the directionality and strength of the mean estimate of the indicator for each latent CIP related to the mean of the population (z-score of 0). This visualization aided in the qualitative labeling of the CIPs.

dc81af7b-4ceb-49f9-9960-8b51102586d8_figure2.gif

Figure 2. Standardized z-score distribution of constrained Contraceptive Ideational Profile indicators by latent status at baseline.

This figure visualizes the directionality and strength of the mean estimate of the indicator for each latent CIP related to the mean of the population (z-score of 0).

CIP 1: Lowest efficacy and FP awareness, highest misconceptions, unsupported

The three most notable attributes for CIP 1 were extremely low self-efficacy, high misconceptions about FP, and low contraceptive awareness. These were the indicators representing the emotional and cognitive domains of contraceptive ideation. Women in CIP 1 also were more likely to report that their partners forbid FP use and did not believe that their communities would support their decisions to use FP. CIP 1 had the smallest membership at baseline (7%) and endline (4%).

CIP 2: Low efficacy and FP awareness, rejects misconceptions, unsupported

In CIP 2, women still had low perceived self-efficacy as the strongest attribute, but it was much less negative than CIP 1. Partner communication, social support, partner support, and contraceptive awareness were all relatively low. One notable positive attribute for this group was their rejection of misconceptions about FP. CIP 2 had the second smallest proportional membership at baseline (12%) and endline (17%).

CIP 3: Moderate efficacy, high FP awareness, high misconceptions, moderate support

In CIP 3 the defining attribute was high misconceptions about FP. Otherwise, women had above-average contraceptive awareness and average efficacy, social, and partner support. CIP 3 was the second largest group at baseline (34%) and endline (19%).

CIP 4: Highest efficacy and FP awareness, fewest misconceptions, most supported

In CIP 4, women had relatively high contraceptive awareness, strongly rejected misconceptions, had high perceived self-efficacy, had good partner support for FP, were more likely to have discussed FP with their partners in the past six months, and were likely to believe that their communities would support their decisions to use an FP method. CIP 4 had the largest membership at both baseline (47%) and endline (60%).

Figure 3 visualizes how women transitioned from their assigned (most likely) CIP at baseline to their endline CIP. The thickness of the lines represents the proportionality with which women transitioned from their CIP at baseline to their CIP at endline. Women who started in baseline CIP 1 transitioned fairly evenly across CIPs 1–4 at endline, with about one-third of the group moving into endline CIP 4. Of women who started in CIPs 2 and 3 at baseline, about half moved to CIP 4 at endline, representing improvement in ideation. In absolute numbers, the largest transition was women who moved from baseline CIP 3 to endline CIP 4 (n=770). Nearly three-quarters of women who started in CIP 4 at baseline stayed in CIP 4 at endline (n=1339). Only a very small proportion of women who started in CIPs 2-4 at baseline transitioned into CIP 1 at endline (n=119).

dc81af7b-4ceb-49f9-9960-8b51102586d8_figure3.gif

Figure 3. Sankey diagram of transitions from baseline CIPs to endline CIPs.

The thickness of the lines visualizes the proportionality with which women transitioned from their assigned (most likely) CIP at baseline to their endline CIP.

Once we identified and labeled latent CIPs and examined how women moved between CIPS over time, we introduced covariates into three multinomial logistic regression models to determine whether these covariates were predictive of endline status membership and transition between profiles over time. For these analyses, we constrained the item-response probabilities to be equal at baseline and endline so that the interpretation of the profiles was the same. Although we were most interested in identifying covariates that increased the odds of transitions to more empowered CIPs over time (CIPs 2, 3, and 4) making CIP 1 the preferred reference category, CIP 1 at baseline and endline had very small membership, and so the more empowered CIPs with larger membership were used as the reference category. We did not test whether covariates differentiated membership or transitions between CIP 2 and CIP 3 because these were qualitatively different groups and neither represented absolute improvement over the other.

In the first model, we ran a multinomial logistic regression to examine whether individual communication modalities predicted endline CIP membership, controlling for key demographic variables (Table 3). We found that hearing about FP from the TV, radio, a religious leader, or a CHW were all associated with significantly lower odds of membership in endline CIP 1 compared to endline CIP 4 and endline CIP 3, indicating that exposure was associated with improved contraceptive ideation. Hearing about FP from TV, radio, a religious leader, or a CHW were all associated with significantly lower odds of membership in endline CIP 1 compared to endline CIPs 4 and 3 indicating that exposure was associated with improved contraceptive ideation. CIP 1 vs. CIPs 3 and 4 are very different profiles and therefore one might expect to see effects given these exposures. It is also interesting to note which exposures differentiate membership in CIPs that are more similar like CIPs 1 and 2 and CIPs 3 and 4. We see that hearing about FP through TV was associated with significantly lower odds of being in endline CIP1 compared with endline CIP 2 and endline CIP 3 with endline CIP 4. Hearing about FP through community conversations with a CHW was also associated with significantly lower odds of membership in endline CIP 1 compared to CIP 2, but a visit from a CHW was not significant in predicting CIP membership. Demographically, a woman’s parity had no association with her odds of CIP membership but higher levels of education were associated with lower odds of being in endline CIPs 1 and 2 compared to CIPs 3 and 4. Residing in a city outside of Dakar was associated with lower odds of membership in CIP 1 vs. CIP 2 and CIP 3 vs. CIP 4. The demographic results were similar in all models and are therefore omitted from Table 4Table 5.

Table 3. Model 1: Individual communication modalities as predictors of endline CIP membership.

Model 1 shares outputs from a multinomial logistic regression to examine whether individual communication modalities predicted endline CIP membership, controlling for key demographic variables.

eCIP 1
[Ref. eC4]
OR (SE)
eCIP 1
[Ref. eC3]
OR (SE)
eCIP 1
[Ref. eC2]
OR (SE)
eCIP 2
[Ref. eC4]
OR (SE)
eCIP 3
[Ref. eC4]
OR (SE)
Heard about FP on TV in past 3 months0.33** (0.10)0.58* (0.21)0.40** (0.17)0.83
(0.37)
0.56** (0.12)
Heard about FP on radio in past 3 months0.53** (0.16)0.49** (0.19)1.09 (0.42)0.49** (0.13)1.01
(0.22)
Heard about FP from religious leader in past 12
months
0.43** (0.13)0.46** (0.15)0.70 (0.21)0.61** (0.13)0.94
(0.24)
Heard about FP from CHW community conversation
in past 12 months
0.34** (0.15)0.32** (0.16)0.51* (0.25)0.67
(0.21)
1.06
(0.32)
Heard about FP from CHW individual visit in past 12
months
1.14
(0.52)
1.19
(0.69)
1.54 (0.80)0.74
(0.27)
0.96
(0.32)
Parity1.03
(0.06)
1.06
(0.07)
1.10 (0.07)0.94
(0.04)
0.97
(0.05)
Primary Education
[Ref. none/Koranic only]
0.43** (0.18)0.49* (0.22)0.62 (0.26)0.69* (0.14)0.86
(0.20)
Secondary Education
[Ref. none/Koranic only]
0.26** (0.22)0.31* (0.27)0.54 (0.49)0.48** (0.19)0.83
(0.19)
City: Guédiawaye [Ref. Dakar]0.20** (0.15)0.42
(0.34)
0.14** (0.11)1.46
(0.65)
0.50** (0.14)
City: Pikine [Ref. Dakar]0.11** (0.10)0.39
(0.37)
0.07** (0.08)1.51
(0.82)
0.29** (0.09)
City: Mbao [Ref. Dakar]1.58
(0.99)
5.26
(3.08)
0.27** (0.19)5.81
(2.56)
0.30** (0.09)
City: Mbour [Ref. Dakar]0.80
(0.37)
2.96
(1.51)
0.30** (0.19)2.66
(1.09)
0.27** (0.07)
City: Kaolack [Ref. Dakar]0.40** (0.21)1.30
(0.70)
0.04** (0.03)10.05* (4.27)0.31** (0.09)

eC: Endline CIP; *p<0.05; **p<0.01

Table 4. Model 2: Total communication modalities as a predictor of endline CIP membership.

Table 4 presents the results of the multinomial logistic regression to determine whether exposure to FP communication through additional modalities predicted the odds of endline profile membership. The results indicate that each additional modality significantly decreased the odds of membership in endline CIP 1 compared to endline CIPs 2, 3, and 4.

eCIP 1
[Ref. eC4]
OR (SE)
eCIP 1
[Ref. eC3]
OR (SE)
eCIP 1
[Ref. eC2]
OR (SE)
eCIP 2
[Ref. eC4]
OR (SE)
eCIP 3
[Ref. eC4]
OR (SE)
Number of communication modalities0.50** (0.06)0.55** (0.07)0.74** (0.10)0.69** (0.05)0.91 (0.06)

eC: Endline CIP; *p<0.05; **p<0.01

Table 5. Model 3: FP message content as predictors of endline CIP membership.

This multinomial logistic regression analysis indicates that only exposure to messages about the legitimacy of FP and birth spacing were significant in predicting more empowered CIP membership at endline.

eCIP 1
[Ref. eC4]
OR (SE)
eCIP 1
[Ref. eC3]
OR (SE)
eCIP 1
[Ref. eC2]
OR (SE)
eCIP 2
[Ref. eC4]
OR (SE)
eCIP 3
[Ref. eC4]
OR (SE)
Legitimacy of FP0.41**
(0.17)
0.57
(0.27)
0.61
(0.29)
0.66*
(0.16)
0.71
(0.16)
Spacing births0.16**
(0.07)
0.24**
(0.11)
0.24**
(0.11)
0.69
(0.22)
0.66*
(0.13)
Spousal communication1.06
(0.67)
0.93
(0.60)
0.58
(0.39)
1.83
(0.59)
1.13
(0.32)
Position of Islam2.11
(1.07)
1.73
(0.89)
2.92
(1.72)
0.72
(0.29)
1.22
(0.36)
Limitation of family size0.39
(0.40)
0.35
(0.37)
0.46
(0.49)
0.85
(0.24)
1.14
(0.31)
Rumors and fears around FP0.71
(0.49)
0.63
(0.44)
1.37
(1.12)
0.52
(0.25)
1.13
(0.37)

eC: Endline CIP; *p<0.05; **p<0.01

In our second model, we summed the total number of unique modalities through which women reported having heard about FP (0–11) to create a count variable of modality breadth to determine whether hearing about FP through additional modalities influenced endline status membership. Table 4 presents the results of the multinomial logistic regression to determine whether exposure to FP communication through additional modalities predicted the odds of endline profile membership. The results indicate that each additional modality significantly decreased the odds of membership in endline CIP 1 compared to endline CIPs 2, 3, and 4. Further, additional modalities significantly decreased the odds of membership in endline CIP 2 compared to endline CIP 4 but was not significant in differentiating endline CIP 3 compared to endline CIP 4.

In our third model, we examined whether the content of the messages that women reported receiving influenced CIP membership. We included six different messages that at least 10 percent of women had reported hearing. Table 5 presents the results of this multinomial logistic regression analysis. Only exposure to messages about the legitimacy of FP and birth spacing were significant in predicting more empowered CIP membership at endline. Women who heard messages about the legitimacy of FP had significantly lower odds of being members of endline CIPs 1 and 2 compared to endline CIP 4. Women who heard messages about birth spacing had significantly lower odds of being in endline CIP 1 compared to endline CIPs 2, 3 and 4 and of being in endline CIP 3 compared to endline CIP 4.

Predictors of transitions between CIPs over time

For each model, we examined whether the significant covariates of interest influenced the odds of transition conditional on baseline CIP membership. In these models, we only included the key covariates of interest that had been significant in the earlier endline models. For each baseline status, we estimated the odds of remaining in that CIP at endline compared to the odds of transitioning to a more empowered endline CIP. Table 6 presents the odds ratios for transitions to endline CIP conditional on baseline CIP for the three models.

Table 6. Predictors of transitions between latent CIPs over time.

Table 6 presents the odds ratios for transitions to endline CIP conditional on baseline CIP for the three models.

Baseline CIP 1Baseline
CIP 2
Baseline
CIP 3
eC1
(Ref. eC2)
OR (SE)
eC1
(Ref. eC3)
OR (SE)
eC1
(Ref. eC4)
OR (SE)
eC2
(Ref. eC4)
OR (SE)
eC3
(Ref. eC4)
OR (SE)
Model 4: Individual Communication Modalities
Heard about FP on TV0.18**
(0.16)
0.69
(0.70)
0.43
(0.32)
0.31
(0.45)
0.30**
(0.14)
Heard about FP on radio4.14
(5.14)
0.17**
(0.18)
0.14**
(0.13)
0.46
(0.64)
0.79
(0.32)
Heard about FP from religious leader1.76
(1.62)
0.29*
(0.32)
0.28**
(0.25)
0.16**
(0.14)
1.49
(0.85)
Heard about FP from CHW0.95
(0.99)
0.96
(1.09)
2.49
(2.24)
0.22**
(0.23)
0.91
(0.42)
Model 5: Total Communication Modalities
Number of communication modalities0.89
(0.30)
0.49**
(0.15)
0.46**
(0.13)
0.34**
(0.14)
0.80*
(0.08)
Model 6: FP Message Content
Legitimacy of FP0.88
(0.85)
0.31*
(0.29)
0.63
(0.54)
0.44
(0.29)
0.55
(0.25)
Spacing births0.44
(0.35)
0.14**
(0.14)
0.15**
(0.10)
0.29*
(0.31)
0.39**
(0.15)

eC: Endline CIP; *p<0.05; **p<0.01

In Model 4, among women who were in baseline CIP 1, exposure to FP content on TV was associated with significantly lower odds of staying in CIP 1 at endline compared to transitioning to endline CIP 2 and hearing about FP on the radio or from a religious leader was associated with significantly lower odds staying in endline CIP 1 compared to transitioning to endline CIP 3 and endline CIP 4. For women who were in baseline CIP 2, hearing about FP from a religious leader or from a CHW was associated with significantly lower odds of staying in CIP 2 at endline compared with transitioning to endline CIP 4. For women who were in baseline CIP 3, only hearing about FP on TV was associated with significantly lower odds of staying in endline CIP 3 compared to transitioning to endline CIP 4.

In Model 5, we examined whether additional modalities influenced the odds of transition conditional on baseline CIP membership. For this model, we only included the total communication modalities covariate for reasons discussed earlier. Results indicated that hearing about FP through additional sources was associated with significantly lower odds of staying in lower CIPs compared to transitions to higher CIPs except from baseline CIP 1 to endline CIP 2.

In Model 6, we examined whether FP message content influenced the odds of transition conditional on baseline CIP membership. We only included content messages about the legitimacy of FP and birth spacing because those were the only two messages that were significant in predicting endline CIP in earlier models. When we conditioned on baseline CIP, the message about the legitimacy of FP was only associated with a significantly lower odds of staying in CIP 1 compared with transition to endline CIP 3. The message about birth spacing was associated with significantly lower odds of staying in a lower CIP at endline compared to transitioning to all more empowered CIPs at endline except between baseline CIP 1 and endline CIP 2.

Discussion

To our knowledge, this study is the first to demonstrate the utility of longitudinal latent variable modeling of contraceptive ideation and its particular relevance to FP communication programs. We identified four prototypical profiles of women who demonstrated unique defining characteristics of contraceptive ideation that indicated that they might be differentially receptive to various FP communication modalities and messages. The four profiles included a group of women who reported very low scores on all six ideational indicators (CIP 1: Lowest efficacy and FP awareness, highest misconceptions, unsupported); two qualitatively different profiles of women, one of which was driven by low self-efficacy but a rejection of misconceptions about FP and the other of which was characterized by strong misconceptions about FP but fairly high contraceptive awareness (CIP 2: Low efficacy and FP awareness, rejects misconceptions, unsupported, and CIP 3: Moderate efficacy, high FP awareness, high misconceptions, moderate support, respectively); and a fourth profile in which women were cognitively, emotionally, and socially empowered in their contraceptive ideation (CIP 4: Highest efficacy and FP awareness, fewest misconceptions, most supported). The insights about these groups of women can help decision-makers appropriately allocate resources to the design of responsive FP interventions and messages. For example, women in CIPs 1 and 2 might benefit from empowering messages that focus on bolstering their self-efficacy whereas women in CIP 3 might benefit from myth-busting positive messaging. Using longitudinal data, we found that over a four-year period (2011–2015) women generally transitioned to become members of more empowered CIPs.

In terms of predictors of endline CIP membership, we found that TV, radio, religious leaders, and community conversations with CHW were all effective modalities in that they each were associated with lower odds of membership in less empowered endline CIPs compared to more empowered CIP 4. Only TV and CHWs differentiated endline membership between endline CIP 1 and endline CIP 2. Another way to interpret this insight is that hearing about FP from TV and from discussion with CHW was a significant differentiator for endline CIP 1 compared to all other CIPs. Communications programs might consider using these channels if the primary goal is to just move people out of CIP 1. Unsurprisingly, exposure to additional communication modalities was also predictive of membership in more empowered endline CIPs. Programs should therefore strive to reinforce FP messages through different channels, as this may increase the credibility, strength, or internalization of the messages for women, their partners, and their communities, all of which can contribute to more positive ideation. In terms of content, we found that messages about the legitimacy of FP and birth spacing were the strongest predictors of membership in more empowered ideational CIPs among the different communication messages used in this program. It was interesting to note that messages about birth spacing were consistent differentiators of endline CIP 1 from all other CIPs, so perhaps this message resonated most with this least empowered group. Programs should consider weaving these themes into their other messages around spousal communication, the position of Islam on FP, the limitation of family size, or rumors and fears about FP.

When we examined transitions that were conditioned on baseline membership, one interesting finding was that the significance and magnitude of many of the indicators changed compared with the more simple endline analyses (Models1–3). This suggests that history matters; that the effects of these exposures are influenced by where a woman is coming from, ideationally.

FP programs can use LTA methodology with nationally representative population data or with smaller programmatic datasets to target women and communities with positive health messaging through appropriate communication channels. Knowing how women in different CIPs access and respond to communication content and modalities can help programs reach the audience segments with effective messages through trusted channels.

Limitations

This study was subject to several limitations. The first limitation is the subjectivity in the selection of the indicators and number of profiles. While we based our decisions on evidence, prior research, and theory, arguments could be made for alternative indicators which could in turn affect the number of profiles selected. For our selected four-CIP LTA model, the entropy score was slightly lower than the desired 0.80 threshold, indicating a greater degree of classification uncertainty. However, for all the covariate model analyses, we used widely accepted three-step BCH method to correct for bias in the estimation of the effect of the covariates on status membership and transition. A third challenge was sparseness due to the large number of cells in the contingency table. Sparseness refers to having a small number of individuals with certain response patterns because there are so many response pattern options, which leads to small or no counts for each cell. This meant that we had to be judicious about the number of covariates to include. Another limitation was that we were not able to control for the total number of exposures in any of the models. For example, we do not know how many times a woman heard about FP on the TV or the radio. Therefore, it is possible that TV was more effective than other modalities because a woman heard about FP multiple times on TV, rather than simply trusting TV as a source that influenced transition.

Conclusion

This study is novel and important for three reasons. First, we demonstrated that latent variable modeling can effectively segment women based on their patterns of social, emotional, and cognitive factors underlying contraceptive ideation. Second, our analysis shed light on women’s transitions between ideational statuses over time. Third, we explored how health communication message content and modality may have influenced those transitions. These insights can be used to create and shape engaging message content and disseminate it through appropriate channels based on ideational needs and channel preferences. These steps can be replicated in other settings using indicators that are locally available and relevant.

The results of this study have important implications for FP communication programs in Senegal and other settings in which socio-cultural norms limit women’s contraceptive decision-making. Communication programs should apply contraceptive ideational profiling as an enhanced SBC approach to achieve a more nuanced understanding of the ideational states and needs of the population and to respond effectively to those needs. Future work should focus on 1) developing a validated toolkit for selecting indicators for contraceptive ideational profiling based on commonly available data sources, 2) creating clear processes for FP communication programs to implement ideational profiling, and 3) empirically evaluating whether latent ideational profiling is better able to reach and engage constituents with relevant content compared with traditional segmentation approaches. Ultimately, this ideational segmentation can be a novel SBC program enhancement to support the Government of Senegal and other countries to reach their FP 2030 Goals.

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 10 May 2024
Comment
Author details Author details
Competing interests
Grant information
Copyright
Download
 
Export To
metrics
Views Downloads
Gates Open Research - -
PubMed Central
Data from PMC are received and updated monthly.
- -
Citations
CITE
how to cite this article
Mangone E, Speizer I, O'Shea N and Hassmiller Lich K. Identification of latent contraceptive ideational profiles among urban women in Senegal: Transitions and implications for family planning programs [version 1; peer review: 2 approved with reservations]. Gates Open Res 2024, 8:37 (https://doi.org/10.12688/gatesopenres.15409.1)
NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.
track
receive updates on this article
Track an article to receive email alerts on any updates to this article.

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 10 May 2024
Comment
Alongside their report, reviewers assign a status to the article:
Approved - the paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations - A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approved - fundamental flaws in the paper seriously undermine the findings and conclusions

Are you a Gates-funded researcher?

If you are a previous or current Gates grant holder, sign up for information about developments, publishing and publications from Gates Open Research.

You must provide your first name
You must provide your last name
You must provide a valid email address
You must provide an institution.

Thank you!

We'll keep you updated on any major new updates to Gates Open Research

Sign In
If you've forgotten your password, please enter your email address below and we'll send you instructions on how to reset your password.

The email address should be the one you originally registered with F1000.

Email address not valid, please try again

You registered with F1000 via Google, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Google account password, please click here.

You registered with F1000 via Facebook, so we cannot reset your password.

To sign in, please click here.

If you still need help with your Facebook account password, please click here.

Code not correct, please try again
Email us for further assistance.
Server error, please try again.