(Last updated on 2021-11-29)

0. Summary

Study question: how does women’s self-reported QoC outcome measures correlate with women’s self-reported process measures (for both experience and technical quality) + structure ?

Take home messages:

Main results

  1. Client’s reported outcome (i.e., very satisfied) positively correlates with communications experiecne consistently across all study countires.
  2. Tehcnical quality (MII4+) is positively correlated only in two.
  3. Structure has no significant correlation.

study sample

PMA CEI clients sample (a systematic but convenient sample) seems to have roughly similar characteristics with modern method users - except in Rajasthan.

Survey questions

Using 5-item likert scales to measure subjective concept (i.e., satisfaction, clarity, and politeness)

1. Study countries and data

Client exit interview implementation results among all SDPs
Survey Total number of SDPs Total number of SDPs that provide FP services Total number of SDPs where CEI was conducted with FP clients Total number of FP clients interviewed Total number of FP clients interviewed who received methods/prescription
Burkina Faso 234 222 77 669 522
DRC, Kinshasa 203 127 10 83 63
DRC, Kongo Central 153 119 13 63 57
India, Rajasthan 574 507 76 394 382
Kenya 945 926 441 3335 3213
Nigeria, Kano 65 54 31 536 500
Nigeria, Lagos 127 112 41 427 332
Uganda 349 333 152 2096 1880
  • Exclude DRC because of small client sample size (both women and facilities) and seemingly very different implementation (substantially low % of SDPs where CEI was conducted).
  • In Burkina Faso and Lagos, Nigeria, less than 80% of FP clients received a method or prescription (i.e., analysis sample).

2. Analysis sample characteristics

Analysis sample = clients who either received a method or prescription

  • Compared to typical profile of modern method users in each country, We expect to see mostly LARC and short acting hormonal methods users (but not male condoms or EC) and any background associated with that methods distribution.
  • Except in Rajasthan, India, age and education distribution among analysis sample is roughly similar with that among modern method users (from FQ).
  • The difference in Rajasthan (i.e., much younger and more educated analysis sample) is explained by different method distributions between the CEI analysis sample and overall modern method users.

2.1 Age

2.1 Education

2.3. Methods

  • Questions/note:
    – Note relatively large share of LARC (basically, IUD + PPIUD) and injectables in Rajasthan CEI sample.
    – 12% of male sterilization in BF - presumably those who received prescription/referral.
    – In Uganda, no distinction between SP & DMPA-IM in CEI. Would be great to separate them in the next round/phase.

modern method users who started using the current methods in the last two years

3. QoC metrics by domain - Levels

  • Largely follows Donabedian’s three domains: structure, process, and outcome

3.1. Outcome

  • Satisfaction reported using a 5-level scale
  • Two other questions (would you return, and would you refer) are binary, and universally high

3.2. Process - Technical

  • Nine questions related with technical competence
    – Eight asked to only analysis sample (Note: no skip pattern (like MII questions in FQ)
    – One asked to all FP clients
  • Various ways to summarize/measure techcnial competence
    – Simple additive score (0-9)
    – categorical variable based on a summary score (e.g., tertiles of scoe based on PCA)
    – Particular/specific components (e.g., MII4+) - more useful for program managers.
    – Bruce/Jaine domains related with technical competence?

For now, MII4 (binary) is used in multivariate regression.

3.3. Process - Experience

  • Information clarity and politeness reported using a 5-level scale
  • Two binary questions on asking quesitons.
  • Again, various ways to summarize/measure experience
    – Particular/specific components (e.g., three communication related questions) - more useful for program managers.
    – Binary summary variable on all four expeerience quesitons.
    – Simple additive score (0-4 or 0-3)

3.4. More on process - Technical and experiential

PCA results show that Technical vs. experiential questions do capture two different dimensions.

In most cases, country-specific PCA results are similar with those using pooled, unweighted data - except in India.

3.5. Structure

  • % of FP CEI analysis sample who facility has met the following criteria:
Client exit interview implementation results among all SDPs
Survey Average number of FP clients interviewed per facility Average number of FP clients interviewed who received methods/prescription Total number of analysis sample
BFP1 8.7 6.8 522
CDKinshasaP1 8.3 6.3 63
CDKongoCentralP1 4.8 4.4 57
INRajasthanP1 5.2 5.0 382
KEP1 7.6 7.3 3213
NGKanoP1 17.3 16.1 500
NGLagosP1 10.4 8.1 332
UGP1 13.8 12.4 1880

4. Multivariate, logistic regression analysis

  • Analysis by country

  • Unit of analysis: FP clients who received methods/prescription

  • Random effect at facility-level (xtlogit outcome covariates, i(facility)). Results are qualitatively similar with those from multilevel mixed-effects generalized linear models (meglm), which was tried per Saifuddin’s suggestion (more on that later…)

  • Outcome: reporting “very satisfied” (a measure for QoC outcome domain)

  • Covariates:

  1. Individual: age (15-24, 25-34, vs. 35+), education (>=attended secondary school), method use status before visit (used same method, used different method, vs. did not use any method), LARC
  2. QOC process, technical: mii4 (binary, yes to all four MII+ items)
    — NOTE: mii9 treid. qualitative similar.
  3. QOC process, experiential: communication_all (binary, yes to all three communication items).
    — NOTE: “very polite” was not included, because it’s subjective/less specific for interventions. But, when included (yes to all four experience items), the results are qualitatively similar.
  4. QOC structure/input: essential5_rnoso (the most restrictive readiness measure)

Adjusted OR of reporting very satisfied, by survey - full model