Introduction
Plausibility check is a tool that evaluates the overall quality and acceptability of anthropometric data to ensure its suitability for informing decision-making process.
mwana provides a set of handy functions to facilitate
this evaluation. These functions allow users to assess the acceptability
of weight-for-height z-score (WFHZ) and mid upper-arm circumference
(MUAC) data. The evaluation of the latter can be done on the basis of
MUAC-for-age z-score (MFAZ) or raw MUAC values.
In this vignette, we will learn how to use these functions and when
to consider using MFAZ plausibility check over the one based on raw MUAC
values. For demonstration, we will use a mwana built-in
sample data set named anthro.01. This data set contains
district level SMART surveys from anonymized locations. Do
?anthro.01 in R console to read more about
it.
We will begin the demonstration with the plausibility check that you are most familiar with and then proceed to the ones you are less familiar with.
Plausibility check of WFHZ data
We check the plausibility of WFHZ data by calling the
mw_plausibility_check_wfhz() function. Before doing that,
we need ensure the data is in the right “shape and format” that is
accepted and understood by the function. Don’t worry, you will soon
learn how to get there. But first, let’s take a moment to walk you
through some key features about this function.
mw_plausibility_check_wfhz() is a replica of the
plausibility check in ENA for SMART software of the SMART Methodology
(SMART Initiative, 2017). Under the hood,
it runs the same test suite you already know from SMART. It also applies
the same rating and scoring criteria. Beware though that there are some
small differences to have in mind:
mw_plausibility_check_wfhz()does not include MUAC in its test suite. This is simply due the fact that now you can run a more comprehensive test suite for MUAC.mw_plausibility_check_wfhz()allows user to run checks on a multiple-area data set at once, without having to repeat the same workflow over and over again for the number of areas the data holds.
That is it! Now we can begin delving into the “how to”.
It is always a good practice to start off by inspecting our data set. Let’s check the first 6 rows of our data set:
head(anthro.01)## # A tibble: 6 × 11
## area dos cluster team sex dob age weight height edema muac
## <chr> <date> <int> <int> <chr> <date> <int> <dbl> <dbl> <chr> <int>
## 1 District E 2023-12-04 1 3 m NA 59 15.6 109. n 146
## 2 District E 2023-12-04 1 3 m NA 8 7.5 68.6 n 127
## 3 District E 2023-12-04 1 3 m NA 19 9.7 79.5 n 142
## 4 District E 2023-12-04 1 3 f NA 49 14.3 100. n 149
## 5 District E 2023-12-04 1 3 f NA 32 12.4 92.1 n 143
## 6 District E 2023-12-04 1 3 f NA 17 9.3 77.8 n 132We can see that the data set has eleven variables, and the way how their respective values are presented. This is useful to inform the data wrangling workflow.
Data wrangling
As mentioned somewhere above, before we supply a data object to
mw_plausibility_check_wfhz(), we need to wrangle it first.
This task is executed by mw_wrangle_age() and
mw_wrangle_wfhz(). Read more about the technical
documentation by doing help("mw_wrangle_age") or
help("mw_wrangle_wfhz") in R console.
Wrangling age
We use mw_wrangle_age() to calculate child’s age in
months based on the date of data collection and child’s date of birth.
This is done as follows:
age_mo <- mw_wrangle_age( # <1>
df = anthro.01 # <2>
dos = dos, # <3>
dob = dob, # <4>
age = age, # <5>
.decimals = 2 # <6>
)The output for this operation will be assigned to an object called
age_mo.The argument
dfis supplied with theanthro.01object which contains variables related to age that will be used for the wrangling process.The argument
dosis supplied with the unquoted variable name indfthat contains the date when the data collection was performed. In theanthro.01dataset, this so happens to bedosas well.The argument
dobis supplied with the unquoted variable name indfthat contains the date when the child was born. In theanthro.01dataset, this so happens to bedobas well.The argument
ageis supplied with the unquoted variable name indfthat contains the age of the child in months. In theanthro.01dataset, this so happens to beageas well.The argument
.decimalsallows the user to specify the number of decimal places to which the output age values will be rounded off to. By default,.decimalsis set to 2. So, even without specifying this argument, the resulting output will be rounded off to 2.
This will return:
## # A tibble: 6 × 12
## area dos cluster team sex dob age weight height edema muac age_days
## <chr> <date> <int> <int> <chr> <date> <int> <dbl> <dbl> <chr> <int> <dbl>
## 1 District E 2023-12-04 1 3 m NA 59 15.6 109. n 146 1796.
## 2 District E 2023-12-04 1 3 m NA 8 7.5 68.6 n 127 244.
## 3 District E 2023-12-04 1 3 m NA 19 9.7 79.5 n 142 578.
## 4 District E 2023-12-04 1 3 f NA 49 14.3 100. n 149 1491.
## 5 District E 2023-12-04 1 3 f NA 32 12.4 92.1 n 143 974
## 6 District E 2023-12-04 1 3 f NA 17 9.3 77.8 n 132 517.Wrangling all other remaining variables
For this, we call mw_wrangle_wfhz() as follows:
wrangled_df <- anthro.01 |>
mw_wrangle_wfhz(
sex = sex,
weight = weight,
height = height,
.recode_sex = TRUE
)In this example, the argument .recode_sex was set to
TRUE. That is because under the hood, to compute the
z-scores, a task made possible thanks to the {zscorer}
package (Myatt and Guevarra, 2019), it
uses sex coded into 1 and 2 for male and female, respectively. This
means that if our sex variable is already in 1 and 2’s, we would set it
to FALSE.
If by any chance your sex variable is coded in any other different
way than aforementioned, then you will have to recode it outside
mwana utilities and then set .recode_sex
accordingly.
Under the hood, after recoding (or not) the sex variables,
mw_wrangle_wfhz() computes the z-scores, then identifies
outliers and adds them to the data set. Two new variables
(wfhz and flag_wfhz) are created and added to
the data set. We can see this below:
## =======================================================================================================## # A tibble: 6 × 3
## area wfhz flag_wfhz
## <chr> <dbl> <dbl>
## 1 District E -1.83 0
## 2 District E -0.956 0
## 3 District E -0.796 0
## 4 District E -0.74 0
## 5 District E -0.679 0
## 6 District E -0.432 0On to de facto plausibility check of WFHZ data
We can check the plausibility of our data by calling
mw_plausibility_check_wfhz() function as demonstrated
below:
x <- wrangled_df |>
mw_plausibility_check_wfhz(
sex = sex,
age = age,
weight = weight,
height = height,
flags = flag_wfhz
)Or we can chain all previous functions in this way:
x <- anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_wfhz(
sex = sex,
weight = weight,
height = height,
.recode_sex = TRUE
) |>
mw_plausibility_check_wfhz(
sex = sex,
age = age,
weight = weight,
height = height,
flags = flag_wfhz
)The returned output is:
## =======================================================================================================## # A tibble: 1 × 19
## n flagged flagged_class sex_ratio sex_ratio_class age_ratio age_ratio_class dps_wgt dps_wgt_class
## <int> <dbl> <fct> <dbl> <chr> <dbl> <chr> <dbl> <chr>
## 1 1191 0.0101 Excellent 0.297 Excellent 0.409 Excellent 3.81 Excellent
## # ℹ 10 more variables: dps_hgt <dbl>, dps_hgt_class <chr>, sd <dbl>, sd_class <chr>, skew <dbl>,
## # skew_class <fct>, kurt <dbl>, kurt_class <fct>, quality_score <dbl>, quality_class <fct>As we can see, the returned output is a summary table of statistics
and ratings. We can neat it for more clarity and readability. We can
achieve this by chaining mw_neat_output_wfhz() to the
previous pipeline:
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_wfhz(
sex = sex,
weight = weight,
height = height,
.recode_sex = TRUE
) |>
mw_plausibility_check_wfhz(
sex = sex,
age = age,
weight = weight,
height = height,
flags = flag_wfhz
) |>
mw_neat_output_wfhz()This will give us:
## =======================================================================================================## # A tibble: 1 × 19
## `Total children` `Flagged data (%)` `Class. of flagged data` `Sex ratio (p)` `Class. of sex ratio`
## <int> <chr> <fct> <chr> <chr>
## 1 1191 1.0% Excellent 0.297 Excellent
## # ℹ 14 more variables: `Age ratio (p)` <chr>, `Class. of age ratio` <chr>, `DPS weight (#)` <dbl>,
## # `Class. DPS weight` <chr>, `DPS height (#)` <dbl>, `Class. DPS height` <chr>,
## # `Standard Dev* (#)` <dbl>, `Class. of standard dev` <chr>, `Skewness* (#)` <dbl>,
## # `Class. of skewness` <fct>, `Kurtosis* (#)` <dbl>, `Class. of kurtosis` <fct>,
## # `Overall score` <dbl>, `Overall quality` <fct>An already formatted table, with scientific notations converted to standard notations, etc.
When working on a multiple-area data set, for instance districts, we
can check the plausibility of all districts in the data set at once by
using group_by() function from the {dplyr}
package as follows:
## Load library ----
library(dplyr)
## The workflow ----
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_wfhz(
sex = sex,
weight = weight,
height = height,
.recode_sex = TRUE
) |>
group_by(area) |>
mw_plausibility_check_wfhz(
sex = sex,
age = age,
weight = weight,
height = height,
flags = flag_wfhz
) |>
group_by(area) |>
mw_neat_output_wfhz()This will return the following:
## =======================================================================================================## # A tibble: 2 × 20
## # Groups: Group [2]
## Group `Total children` `Flagged data (%)` `Class. of flagged data` `Sex ratio (p)`
## <chr> <int> <chr> <fct> <chr>
## 1 District E 505 0.8% Excellent 0.593
## 2 District G 686 1.2% Excellent 0.380
## # ℹ 15 more variables: `Class. of sex ratio` <chr>, `Age ratio (p)` <chr>,
## # `Class. of age ratio` <chr>, `DPS weight (#)` <dbl>, `Class. DPS weight` <chr>,
## # `DPS height (#)` <dbl>, `Class. DPS height` <chr>, `Standard Dev* (#)` <dbl>,
## # `Class. of standard dev` <chr>, `Skewness* (#)` <dbl>, `Class. of skewness` <fct>,
## # `Kurtosis* (#)` <dbl>, `Class. of kurtosis` <fct>, `Overall score` <dbl>, `Overall quality` <fct>At this point, you have reached the end of your workflow 🎉 .
Plausibility check of MFAZ data
We will assess the plausibility of MUAC data through MFAZ if we have age variable available in our data set.
The plausibility check for MFAZ data was built based on the insights gotten from Bilukha and Kianian (2023) research presented at the 2023 High-Level Technical Assessment Workshop held in Nairobi, Kenya (SMART Initiative, 2023). Results from this research suggested a feasibility of applying the similar plausibility check as that of WFHZ for MFAZ, with a maximum acceptability of percent of flagged records of 2.0%.
We can run MFAZ plausibility check by calling
mw_plausibility_check_mfaz(). As in WFHZ, we first need to
ensure that the data is in the right shape and format that is accepted
and understood by the function. The workflow starts with wrangling age;
for this, we approach the same way as in (sec-age?).
Age ratio test in MFAZ
As you know, the age ratio test in WFHZ is done on children aged 6 to 29 months old over those aged 30 to 59 months old. This is different in MFAZ. The test is done on children aged 6 to 23 months over those aged 24 to 59 months old. This is as in the SMART MUAC Tool (SMART Initiative, n.d.). The test results is also used in the prevalence analysis to implement what the SMART MUAC tool does. This is further demonstrated in the vignette about prevalence.
Wrangling MFAZ data
This is the job of mw_wrangle_muac() function. We use it
as follows:
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = "age",
.recode_sex = TRUE,
.recode_muac = TRUE,
.to = "cm"
)Just as in WFHZ wrangler, under the hood,
mw_wrangle_muac() computes the z-scores then identifies
outliers and flags them. These are stored in the mfaz and
flag_mfaz variables that are created and added to the data
set.
The above code returns:
## =======================================================================================================## # A tibble: 1,191 × 14
## area dos cluster team sex dob age weight height edema muac age_days mfaz
## <chr> <date> <int> <int> <dbl> <date> <int> <dbl> <dbl> <chr> <dbl> <dbl> <dbl>
## 1 District E 2023-12-04 1 3 1 NA 59 15.6 109. n 14.6 1796. -1.45
## 2 District E 2023-12-04 1 3 1 NA 8 7.5 68.6 n 12.7 244. -1.67
## 3 District E 2023-12-04 1 3 1 NA 19 9.7 79.5 n 14.2 578. -0.617
## 4 District E 2023-12-04 1 3 2 NA 49 14.3 100. n 14.9 1491. -1.02
## 5 District E 2023-12-04 1 3 2 NA 32 12.4 92.1 n 14.3 974 -0.93
## 6 District E 2023-12-04 1 3 2 NA 17 9.3 77.8 n 13.2 517. -1.10
## 7 District E 2023-12-04 1 3 2 NA 20 10.1 80.4 n 14.3 609. -0.255
## 8 District E 2023-12-04 1 3 2 NA 27 11.7 87.1 n 14.3 822. -0.677
## 9 District E 2023-12-04 1 3 1 NA 46 13.6 98 n 13.5 1400. -2.18
## 10 District E 2023-12-04 1 3 1 NA 58 17.2 109. n 15.9 1765. -0.403
## # ℹ 1,181 more rows
## # ℹ 1 more variable: flag_mfaz <dbl>mw_wrangle_muac() accepts MUAC values in centimeters.
This is why it takes the arguments .recode_muac and
.to to control whether there is need to transform the
variable muac or not. Read the function documentation to
learn about how to control these two arguments.
On to de facto plausibility check of MFAZ data
We achieve this by calling the
mw_plausibility_check_mfaz() function:
## Load dplyr library ----
library(dplyr)
## The workflow ----
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = "age",
.recode_sex = TRUE,
.recode_muac = TRUE,
.to = "cm"
) |>
mutate(muac = recode_muac(muac, .to = "mm")) |>
mw_plausibility_check_mfaz(
sex = sex,
muac = muac,
age = age,
flags = flag_mfaz
)And this will return:
## =======================================================================================================## # A tibble: 1 × 17
## n flagged flagged_class sex_ratio sex_ratio_class age_ratio age_ratio_class dps dps_class sd
## <int> <dbl> <fct> <dbl> <chr> <dbl> <chr> <dbl> <chr> <dbl>
## 1 1191 0.00504 Excellent 0.297 Excellent 0.636 Excellent 100 Problema… 0.767
## # ℹ 7 more variables: sd_class <chr>, skew <dbl>, skew_class <fct>, kurt <dbl>, kurt_class <fct>,
## # quality_score <dbl>, quality_class <fct>We can also neat this output. We just need to call
mw_neat_output_mfaz() and chain it to the pipeline:
## Load dplyr library ----
library(dplyr)
## The workflow ----
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = "age",
.recode_sex = TRUE,
.recode_muac = TRUE,
.to = "cm"
) |>
mutate(muac = recode_muac(muac, .to = "mm")) |>
mw_plausibility_check_mfaz(
sex = sex,
muac = muac,
age = age,
flags = flag_mfaz
) |>
mw_neat_output_mfaz()This will return:
## =======================================================================================================## # A tibble: 1 × 17
## `Total children` `Flagged data (%)` `Class. of flagged data` `Sex ratio (p)` `Class. of sex ratio`
## <int> <chr> <fct> <chr> <chr>
## 1 1191 0.5% Excellent 0.297 Excellent
## # ℹ 12 more variables: `Age ratio (p)` <chr>, `Class. of age ratio` <chr>, `DPS (#)` <dbl>,
## # `Class. of DPS` <chr>, `Standard Dev* (#)` <dbl>, `Class. of standard dev` <chr>,
## # `Skewness* (#)` <dbl>, `Class. of skewness` <fct>, `Kurtosis* (#)` <dbl>,
## # `Class. of kurtosis` <fct>, `Overall score` <dbl>, `Overall quality` <fct>We can also run checks on a multiple-area data set as follows:
## Load dplyr library ----
library(dplyr)
## The workflow ----
anthro.01 |>
mw_wrangle_age(
dos = dos,
dob = dob,
age = age,
.decimals = 2
) |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = "age",
.recode_sex = TRUE,
.recode_muac = TRUE,
.to = "cm"
) |>
mutate(muac = recode_muac(muac, .to = "mm")) |>
group_by(area) |>
mw_plausibility_check_mfaz(
sex = sex,
muac = muac,
age = age,
flags = flag_mfaz
) |>
group_by(area) |>
mw_neat_output_mfaz()This will return:
## =======================================================================================================## # A tibble: 2 × 18
## # Groups: Group [2]
## Group `Total children` `Flagged data (%)` `Class. of flagged data` `Sex ratio (p)`
## <chr> <int> <chr> <fct> <chr>
## 1 District E 505 0.0% Excellent 0.593
## 2 District G 686 0.9% Excellent 0.380
## # ℹ 13 more variables: `Class. of sex ratio` <chr>, `Age ratio (p)` <chr>,
## # `Class. of age ratio` <chr>, `DPS (#)` <dbl>, `Class. of DPS` <chr>, `Standard Dev* (#)` <dbl>,
## # `Class. of standard dev` <chr>, `Skewness* (#)` <dbl>, `Class. of skewness` <fct>,
## # `Kurtosis* (#)` <dbl>, `Class. of kurtosis` <fct>, `Overall score` <dbl>, `Overall quality` <fct>At this point, you have reached the end of your workflow ✨.
Plausibility check of raw MUAC data
We will assess the plausibility of raw MUAC data through it’s raw
values when the variable age is not available in the data set. This is a
job assigned to mw_plausibility_check_muac(). The workflow
for this check is the shortest one.
Data wrangling
As you can tell, z-scores cannot be computed in the absence of age.
In this way, the data wrangling workflow would be quite minimal. You
still set the arguments inside mw_wrangle_muac() as learned
in (sec-wrangle_mfaz?).
The only difference is that here we will set age to
NULL. Fundamentally, under the hood the function detects
MUAC values that are outliers and flags them and stores them in
flag_muac variable that is added to the data set.
We will continue using the same data set:
anthro.01 |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = NULL,
.recode_sex = TRUE,
.recode_muac = FALSE,
.to = "none"
)This returns:
## # A tibble: 1,191 × 12
## area dos cluster team sex dob age weight height edema muac flag_muac
## <chr> <date> <int> <int> <dbl> <date> <int> <dbl> <dbl> <chr> <int> <dbl>
## 1 District E 2023-12-04 1 3 1 NA 59 15.6 109. n 146 0
## 2 District E 2023-12-04 1 3 1 NA 8 7.5 68.6 n 127 0
## 3 District E 2023-12-04 1 3 1 NA 19 9.7 79.5 n 142 0
## 4 District E 2023-12-04 1 3 2 NA 49 14.3 100. n 149 0
## 5 District E 2023-12-04 1 3 2 NA 32 12.4 92.1 n 143 0
## 6 District E 2023-12-04 1 3 2 NA 17 9.3 77.8 n 132 0
## 7 District E 2023-12-04 1 3 2 NA 20 10.1 80.4 n 143 0
## 8 District E 2023-12-04 1 3 2 NA 27 11.7 87.1 n 143 0
## 9 District E 2023-12-04 1 3 1 NA 46 13.6 98 n 135 0
## 10 District E 2023-12-04 1 3 1 NA 58 17.2 109. n 159 0
## # ℹ 1,181 more rowsOn to de facto plausibility check
We just have to add mw_plausibility_check_muac() to the
above pipeline:
anthro.01 |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = NULL,
.recode_sex = TRUE,
.recode_muac = FALSE,
.to = "none"
) |>
mw_plausibility_check_muac(
sex = sex,
flags = flag_muac,
muac = muac
)And this will return:
## # A tibble: 1 × 9
## n flagged flagged_class sex_ratio sex_ratio_class dps dps_class sd sd_class
## <int> <dbl> <fct> <dbl> <chr> <dbl> <chr> <dbl> <fct>
## 1 1191 0.00252 Excellent 0.297 Excellent 5.39 Excellent 11.1 ExcellentWe can also return a formatted table with
mw_neat_output_muac():
anthro.01 |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = NULL,
.recode_sex = TRUE,
.recode_muac = FALSE,
.to = "none"
) |>
mw_plausibility_check_muac(
sex = sex,
flags = flag_muac,
muac = muac
) |>
mw_neat_output_muac()And we get:
## # A tibble: 1 × 9
## `Total children` `Flagged data (%)` `Class. of flagged data` `Sex ratio (p)` `Class. of sex ratio`
## <int> <chr> <fct> <chr> <chr>
## 1 1191 0.3% Excellent 0.297 Excellent
## # ℹ 4 more variables: `DPS(#)` <dbl>, `Class. of DPS` <chr>, `Standard Dev* (#)` <dbl>,
## # `Class. of standard dev` <fct>When working on multiple-area data, we approach the task the same way as demonstrated above:
## Load library ----
library(dplyr)
## Check plausibility ----
anthro.01 |>
mw_wrangle_muac(
sex = sex,
muac = muac,
age = NULL,
.recode_sex = TRUE,
.recode_muac = FALSE,
.to = "none"
) |>
group_by(area) |>
mw_plausibility_check_muac(
sex = sex,
flags = flag_muac,
muac = muac
) |>
group_by(area) |>
mw_neat_output_muac()And we get:
## # A tibble: 2 × 10
## # Groups: Group [2]
## Group `Total children` `Flagged data (%)` `Class. of flagged data` `Sex ratio (p)`
## <chr> <int> <chr> <fct> <chr>
## 1 District E 505 0.0% Excellent 0.593
## 2 District G 686 0.4% Excellent 0.380
## # ℹ 5 more variables: `Class. of sex ratio` <chr>, `DPS(#)` <dbl>, `Class. of DPS` <chr>,
## # `Standard Dev* (#)` <dbl>, `Class. of standard dev` <fct>