In the recent past, measurement of coverage has been mainly through two-stage cluster sampled surveys either as part of a nutrition assessment or through a specific coverage survey known as Centric Systematic Area Sampling (CSAS). However, such methods are resource intensive and often only used for final programme evaluation meaning results arrive too late for programme adaptation. SLEAC, which stands for Simplified Lot Quality Assurance Sampling Evaluation of Access and Coverage, is a low resource method designed specifically to address this limitation and is used regularly for monitoring, planning and importantly, timely improvement to programme quality, both for agency and Ministry of Health (MoH) led programmes. This package provides functions for use in conducting a SLEAC assessment.
The {sleacr} package provides functions that facilitate the design,
sampling, data collection, and data analysis of a SLEAC survey. The
current version of the {sleacr} package currently provides the
following:
-
Functions to calculate the sample size needed for a SLEAC survey;
-
Functions to draw a stage 1 sample for a SLEAC survey;
-
Functions to classify coverage;
-
Functions to determine the performance of chosen classifier cut-offs for analysis of SLEAC survey data;
-
Functions to estimate coverage over wide areas; and,
-
Functions to test for coverage homogeneity across multiple surveys over wide areas.
The {sleacr} package is not yet available on
CRAN but can be installed from the
nutriverse R Universe as follows:
install.packages(
"sleacr",
repos = c('https://nutriverse.r-universe.dev', 'https://cloud.r-project.org')
)To setup an LQAS sampling frame, a target sample size is first estimated. For example, if the survey area has an estimated population of about 600 severe acute malnourished (SAM) children and you want to assess whether coverage is reaching at least 50%, the sample size can be calculated as follows:
get_sample_n(N = 600, dLower = 0.5, dUpper = 0.8)which gives an LQAS sampling plan list with values for the target
minimum sample size (n), the decision rule (d), the observed alpha
error (alpha), and the observed beta error (beta).
#> $n
#> [1] 19
#>
#> $d
#> [1] 12
#>
#> $alpha
#> [1] 0.06446194
#>
#> $beta
#> [1] 0.08014249
In this sampling plan, a target minimum sample size of 19 SAM cases should be aimed for with a decision rule of more than 12 SAM cases covered to determine whether programme coverage is at least 50% with alpha and beta errors no more than 10%. The alpha and beta errors requirement is set at no more than 10% by default. This can be made more precise by setting alpha and beta errors less than 10%.
There are contexts where survey data has already been collected and the
sample is less than what was aimed for based on the original sampling
frame. The get_sample_d() function is used to determine the error
levels of the achieved sample size. For example, if the survey described
above only achieved a sample size of 16, the get_sample_d() function
can be used as follows:
get_sample_d(N = 600, n = 16, dLower = 0.5, dUpper = 0.8)which gives an alternative LQAS sampling plan based on the achieved sample size.
#> $n
#> [1] 16
#>
#> $d
#> [1] 10
#>
#> $alpha
#> [1] 0.07890285
#>
#> $beta
#> [1] 0.1019738
In this updated sampling plan, the decision rule is now more than 10 SAM cases but with higher alpha and beta errors. Note that the beta error is now slightly higher than 10%.
The first stage sample of a SLEAC survey is a systematic spatial sample.
Two methods can be used and both methods take the sample from all parts
of the survey area: the list-based method and the map-based method.
The {sleacr} package currently supports the implementation of the
list-based method.
In the list-based method, communities to be sampled are selected
systematically from a complete list of communities in the survey area.
This list of communities should sorted by one or more non-overlapping
spatial factors such as district and subdistricts within districts. The
village_list dataset is an example of such a list.
village_list
#> # A tibble: 1,001 × 4
#> id chiefdom section village
#> <dbl> <chr> <chr> <chr>
#> 1 1 Badjia Damia Ngelehun
#> 2 2 Badjia Damia Gondama
#> 3 3 Badjia Damia Penjama
#> 4 4 Badjia Damia Jawe
#> 5 5 Badjia Damia Dambala
#> 6 6 Badjia Fallay Bumpewo
#> 7 7 Badjia Fallay Pelewahun
#> 8 8 Badjia Fallay Pendembu
#> 9 9 Badjia Kpallay Jokibu
#> 10 10 Badjia Kpallay Kpaku
#> # ℹ 991 more rowsThe get_sampling_list() function implements the list-based sampling
method. For example, if 40 clusters/villages are needed to be sampled to
find the 19 SAM cases calculated earlier, a sampling list can be created
as follows:
get_sampling_list(village_list, 40)which provides the following sampling list:
| id | chiefdom | section | village |
|---|---|---|---|
| 20 | Badjia | Njargbahun | Kpetema |
| 45 | Bagbe | Jongo | Yengema |
| 70 | Bagbe | Samawa | Baiama |
| 95 | Bagbo | Jimmi | Kpawama |
| 120 | Bagbo | Mano | Dandabu |
| 145 | Baoma | Bambawo | Kenemawo |
| 170 | Baoma | Fallay | Gbandi |
| 195 | Baoma | Mawojeh | Ngelahun |
| 220 | Baoma | Upper Pataloo | Yakaji |
| 245 | Bumpe Ngao | Bumpe | Waiima |
| 270 | Bumpe Ngao | Foya | Bobobu |
| 295 | Bumpe Ngao | Bongo | Belebu |
| 320 | Bumpe Ngao | Serabu | Nyahagoihun |
| 345 | Bumpe Ngao | Taninahun | Kpetewoma |
| 370 | Bumpe Ngao | Taninahun | Mokebi |
| 395 | Bumpe Ngao | Taninahun | Ngiegboiya |
| 420 | Gbo | Gbo | Kotumahun Mavi |
| 445 | Gbo | Nyawa | Foya |
| 470 | Jaiama Bongor | Lower Niawa | Baraka |
| 495 | Jaiama Bongor | Tongowa | Talia |
| 520 | Jaiama Bongor | Upper Niawa | Nyeyama |
| 545 | Kakua | Kpandobu | Fabaina |
| 570 | Kakua | Nyallay | Jandama |
| 595 | Kakua | Sewa | Kenedeyama |
| 620 | Komboya | Kemoh | Gumahun |
| 645 | Komboya | Mangaru | Sengbehun |
| 670 | Lugbu | Kargbevu | Momandu |
| 695 | Niawa Lenga | Lower Niawa | Luawa |
| 720 | Niawa Lenga | Yalenga | Dandabu |
| 745 | Selenga | Mokpendeh | Jolu |
| 770 | Tikonko | Ngolamajie | Baoma (Geyewoma) |
| 795 | Tikonko | Seiwa | Gendema |
| 820 | Tikonko | Seiwa | Towama |
| 845 | Tikonko | Seiwa | Kpawugbahun |
| 870 | Valunia | Deilenga | Hendogboma |
| 895 | Valunia | Lower Kargoi | Gombu |
| 920 | Valunia | Lunia | Kpetema |
| 945 | Valunia | Manyeh | Malema |
| 970 | Valunia | Yarlenga | Dassamu |
| 995 | Wonde | Manyeh | Kigbema |
With data collected from a SLEAC survey, the lqas_classify_coverage()
function is used to classify coverage. The {sleacr} package comes with
the survey_data dataset from a national SLEAC survey conducted in
Sierra Leone.
survey_data
#> # A tibble: 14 × 7
#> country province district cases_in cases_out rec_in cases_total
#> <chr> <chr> <chr> <int> <int> <int> <int>
#> 1 Sierra Leone Northern Bombali 4 26 6 30
#> 2 Sierra Leone Northern Koinadugu 0 32 6 32
#> 3 Sierra Leone Northern Kambia 0 28 0 28
#> 4 Sierra Leone Northern Port Loko 2 28 0 30
#> 5 Sierra Leone Northern Tonkolili 1 27 5 28
#> 6 Sierra Leone Eastern Kono 2 14 3 16
#> 7 Sierra Leone Eastern Kailahun 4 30 3 34
#> 8 Sierra Leone Eastern Kenema 8 26 4 34
#> 9 Sierra Leone Southern Pujehun 6 21 1 27
#> 10 Sierra Leone Southern Bo 6 16 8 22
#> 11 Sierra Leone Southern Bonthe 7 34 2 41
#> 12 Sierra Leone Southern Moyamba 6 34 0 40
#> 13 Sierra Leone Western Area Western Area… 6 40 5 46
#> 14 Sierra Leone Western Area Western Area… 2 18 0 20Using this dataset, per district coverage classifications can be calculated as follows:
with(
survey_data,
lqas_classify(
cases_in = cases_in, cases_out = cases_out, rec_in = rec_in
)
)which outputs the following results:
#> cf tc
#> 1 0 1
#> 2 0 0
#> 3 0 0
#> 4 0 0
#> 5 0 0
#> 6 0 1
#> 7 0 0
#> 8 1 1
#> 9 1 1
#> 10 1 1
#> 11 0 0
#> 12 0 0
#> 13 0 0
#> 14 0 0
The function provides estimates for case-finding effectiveness and for
treatment coverage as a data.frame object.
It is useful to be able to assess the performance of the classifier chosen for a SLEAC survey. For example, in the context presented above of an area with a population of 600, a sample size of 40 and a 60% and 90% threshold classifier, the performance of this classifier can be assessed by first simulating a population and then determining the classification probabilities of the chosen classifier on this population.
## Simulate population ----
lqas_sim_pop <- lqas_simulate_test(
pop = 600, n = 40, dLower = 0.6, dUpper = 0.9
)
## Get classification probabilities ----
lqas_get_class_prob(lqas_sim_pop)
#> Low : 0.9551
#> Moderate : 0.8332
#> High : 0.835
#> Overall : 0.9065
#> Gross misclassification : 0This diagnostic test can also be plotted.
plot(lqas_sim_pop)
When SLEAC is implemented in several service delivery units, it is also
possible to estimate an overall coverage across these service delivery
units. For example, using the survey_data dataset from a national
SLEAC survey conducted in Sierra Leone, an overall coverage estimate can
be calculated. For this, additional information on the total population
for each service delivery unit surveyed will be needed. For the Sierra
Leone example, the pop_data dataset gives the population for each
district in Sierra Leone.
pop_data
#> # A tibble: 14 × 2
#> district pop
#> <chr> <dbl>
#> 1 Kailahun 526379
#> 2 Kenema 609891
#> 3 Kono 506100
#> 4 Bombali 606544
#> 5 Kambia 345474
#> 6 Koinadugu 409372
#> 7 Port Loko 615376
#> 8 Tonkolili 531435
#> 9 Bo 575478
#> 10 Bonthe 200781
#> 11 Moyamba 318588
#> 12 Pujehun 346461
#> 13 Western Area Rural 444270
#> 14 Western Area Urban 1055964The overall coverage estimate can be calculated as follows:
pop_df <- pop_data |>
setNames(nm = c("strata", "pop"))
estimate_coverage_overall(
survey_data, pop_data, strata = "district", u5 = 0.177, p = 0.01
)which gives the following results:
#> $cf
#> $cf$estimate
#> [1] 0.1257481
#>
#> $cf$ci
#> [1] 0.09247579 0.15902045
#>
#>
#> $tc
#> $tc$estimate
#> [1] 0.1706466
#>
#> $tc$ci
#> [1] 0.1371647 0.2041284
When estimating coverage across multiple surveys over wide areas, it is
good practice to assess whether coverage across each of the service
delivery units is homogenous. The function
check_coverage_homogeneity() is used for this purpose:
check_coverage_homogeneity(survey_data)which results in the following output:
#> ℹ Case-finding effectiveness across 14 surveys is not patchy.
#> ! Treatment coverage across 14 surveys is patchy.
#> $cf
#> $cf$statistic
#> [1] 20.1292
#>
#> $cf$df
#> [1] 13
#>
#> $cf$p
#> [1] 0.09203514
#>
#>
#> $tc
#> $tc$statistic
#> [1] 33.10622
#>
#> $tc$df
#> [1] 13
#>
#> $tc$p
#> [1] 0.001642536
In this example, case-finding effectiveness is homogeneous while treatment coverage is patchy.
If you use {sleacr} in your work, please cite using the suggested
citation provided by a call to the citation function as follows:
citation("sleacr")
#> To cite sleacr in publications use:
#>
#> Mark Myatt, Ernest Guevarra, Lionella Fieschi, Allison Norris, Saul
#> Guerrero, Lilly Schofield, Daniel Jones, Ephrem Emru, Kate Sadler
#> (2012). _Semi-Quantitative Evaluation of Access and Coverage
#> (SQUEAC)/Simplified Lot Quality Assurance Sampling Evaluation of
#> Access and Coverage (SLEAC) Technical Reference_. FHI 360/FANTA,
#> Washington, DC.
#> <https://www.fantaproject.org/sites/default/files/resources/SQUEAC-SLEAC-Technical-Reference-Oct2012_0.pdf>.
#>
#> A BibTeX entry for LaTeX users is
#>
#> @Book{,
#> title = {Semi-Quantitative Evaluation of Access and Coverage ({SQUEAC})/Simplified Lot Quality Assurance Sampling Evaluation of Access and Coverage ({SLEAC}) Technical Reference},
#> author = {{Mark Myatt} and {Ernest Guevarra} and {Lionella Fieschi} and {Allison Norris} and {Saul Guerrero} and {Lilly Schofield} and {Daniel Jones} and {Ephrem Emru} and {Kate Sadler}},
#> year = {2012},
#> publisher = {FHI 360/FANTA},
#> address = {Washington, DC},
#> url = {https://www.fantaproject.org/sites/default/files/resources/SQUEAC-SLEAC-Technical-Reference-Oct2012_0.pdf},
#> }Feedback, bug reports, and feature requests are welcome; file issues or seek support here. If you would like to contribute to the package, please see our contributing guidelines.
This project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.