MVDHS data pooling pre-checks

Getting started

Here we show the pre-requisite code sections. Run these at the outset to avoid errors. First we load the required packages.

easypackages::libraries(
  # Data i/o
  "here",                 # relative file path
  "rio",                  # file import-export
  
  # Data manipulation
  "janitor",              # data cleaning fns
  "haven",                # stata, sas, spss data io
  "labelled",             # var labelling
  "readxl",               # excel sheets
  # "scales",               # to change formats and units
  "skimr",                # quick data summary
  "broom",                # view model results
  
  # Data analysis
  "DHS.rates",            # demographic rates for dhs-like surveys
  "GeneralOaxaca",        # BO decomposition for non-linear
  "survey",               # apply survey weights
  
  # Analysis output
  "gt",
  # "modelsummary",          # output summary tables
  "gtsummary",            # output summary tables
  "flextable",            # creating tables from objects
  "officer",              # editing in office docs
  
  # R graph related packages
  "ggstats",
  "RColorBrewer",
  # "scales",
  "patchwork",
  
  # Misc packages
  "tidyverse",            # Data manipulation iron man
  "tictoc"                # Code timing
)

Next we turn off scientific notations.

options(scipen = 999)

Next we set the default gtsummary print engine for tables.

theme_gtsummary_printer(print_engine = "flextable")

Now we set the flextable output defaults.

set_flextable_defaults(
  font.size = 11,
  text.align = "left",
  big.mark = "",
  background.color = "white",
  table.layout = "autofit",
  theme_fun = theme_vanilla
)

Document introduction

Here we document the variable codes and labels of variables across all the Maldives Demographic and Health Survey (DHS) datasets. We check the variable labels and codes before running the pooling code in “daprep-v01_mvdhs.R”. We pool the following Maldives DHS surveys:

# Creating the table of surveys to be used for pooling
mvbr1_tmp_intro |> 
  mutate(n_births = prettyNum(n_births, big.mark = ",")) |> 
  select(c(ctr_name, svy_year, n_births)) |> 
  # Join vars from mvir_tmp_intro
  left_join(
    mvir1_tmp_intro |> 
      mutate(n_women = prettyNum(n_women, big.mark = ",")) |> 
      select(c(year, n_women)),
    by = join_by(svy_year == year)
  ) |> 
  # Join vars from mvhr_tmp_intro
  left_join(
    mvhr1_tmp_intro |> 
      mutate(n_households = prettyNum(n_households, big.mark = ",")) |> 
      select(svy_year, n_households),
    by = join_by(svy_year)
  ) |> 
  # Join vars from mvpr_tmp_intro
  left_join(
    mvpr1_tmp_intro |> 
      mutate(n_persons = prettyNum(n_persons, big.mark = ",")) |> 
      select(svy_year, n_persons),
    by = join_by(svy_year)
  ) |> 
  # convert nested tibble to simple tibble
  unnest(cols = c()) |> 
  mutate(
    ccode = row_number(), 
    .before = ctr_name
  ) |> 
  # convert to flextable object
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 1: Maldives DHS datasets and their sample size to be used for pooling

ccode	ctr_name	svy_year	n_births	n_women	n_households	n_persons
1	Maldives	2009	20,136	7,131	6,443	42,050
2	Maldives	2016	13,922	7,699	6,050	32,656

We use the following variables for the pooled data analysis:

Dependent variable
- infantd = Index child died during infancy period (0-11 months)
Main Independent variable
- sibsurv_nmv = Survival status of preceding child (Death scarring)
- binterval_3c_nmv_opp = Birth interval preceding to index child
Independent variables [CHILD LEVEL]
- cyob10y_opp = Birth cohort of index child
- bord_c = Birth order of index child
- sex_fm = Gender of index child
- season = Season during birth
Independent variables [MOTHER/PARENT LEVEL]
- ~~myob_opp = Birth cohort of mother~~
- macb_c_opp = Mother’s age during birth of index child
- medu_opp = Mother’s Level of education
- fedu_opp = Father’s level of education
Independent variables [HOUSEHOLD LEVEL]
- religion = Religion
- nat_lang = Native language of respondent
- wi_qt_opp = Household wealth quintile
- ~~hhgen_2c_opp = Generations in household~~
- hhstruc_opp = Household structure
- head_sex_fm = Sex of HH head
Independent variables [COMMUNITY LEVEL]
- por = Place of residence of the household
- ecoreg = Ecological region

Note: (a) Crossed names indicates variable not included.

Data import

We will directly import the nested tibble here. The code for dataset preparation is in the “daprep-v01_mvdhs.R” script file.

# Here we import the tibbles to be used for dataset checking
# Import the mvbr nested tibble
mvbr1_pre_tmp0 <- read_rds(file = here("website_data", "mvbr1_nest0.rds"))
# Import the mvhr nested tibble
mvhr1_pre_tmp0 <- read_rds(file = here("website_data", "mvhr1_nest0.rds"))
# Import the mvpr nested tibble
mvpr1_pre_tmp0 <- read_rds(file = here("website_data", "mvpr1_nest0.rds"))

Maldives BR dataset use for variable creation

Checking the Women’s weight variable before harmonization

We will check the formatting of the v005 women’s weight variable before creating the pooled survey weight. For this we will use the labelled::look_for().

# First we create the data dictionary of v005 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v005 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v005) |> 
        look_for(details = "full") |> 
        # For correctly viewing the range column in data dictionary
        convert_list_columns_to_character() |> 
        select(-c(levels:n_na))
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary 
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v005)) |> 
  select(-pos) 
# Convert and view the tibble as flextable
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 2: Data dictionary of v005 variable across the mvbr rounds

ctr_name	svy_year	variable	label	col_type	missing	unique_values	range
Maldives	2009	v005	sample weight	dbl	0	257	99107 - 4890879
Maldives	2016	v005	women's individual sample weight (6 decimals)	dbl	0	173	162566 - 5001387

The women’s weight variables are in numeric class and have no missing values. Therefore, we need not reformat them. Hence we directly use it for preparing the pooled survey weight.

Checking the ID variables before harmonization

Here we check the formatting of the variables using which we will prepare the ID variables for the pooled Maldives birth history recode (br) dataset. We will use the following constituent variables for creating the ID variables for the pooled dataset:

# We check the var type of ID vars in all mvbr datasets.
# First we create a data dictionary of the mvbr datasets in nested tibble.
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |>
  mutate(lookfor_idvars = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v001, v002, v003, bord, v021, v022, v023, v024) |> 
        lookfor(details = "full") |> 
        select(-c(levels:n_na)) |> 
        # For correctly viewing the range column in data dictionary
        convert_list_columns_to_character()
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and output the pooled data dictionary 
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_idvars)) |> 
  arrange(pos)

# Convert and view the tibble as flextable
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 3: Data dictionary of variables to be used for ID creation across the mvbr rounds

ctr_name	svy_year	pos	variable	label	col_type	unique_values	range
Maldives	2009	1	v001	cluster number	dbl	270	1 - 270
Maldives	2016	1	v001	cluster number	dbl	266	1 - 266
Maldives	2009	2	v002	household number	dbl	98	1 - 99
Maldives	2016	2	v002	household number	dbl	41	1 - 42
Maldives	2009	3	v003	respondent's line number	dbl	21	1 - 23
Maldives	2016	3	v003	respondent's line number	dbl	20	1 - 27
Maldives	2009	4	bord	birth order number	dbl	14	1 - 14
Maldives	2016	4	bord	birth order number	dbl	13	1 - 13
Maldives	2009	5	v021	primary sampling unit	dbl	270	1 - 270
Maldives	2016	5	v021	primary sampling unit	dbl	266	1 - 266
Maldives	2009	6	v022	sample stratum number	dbl+lbl	21	1 - 21
Maldives	2016	6	v022	sample strata for sampling errors	dbl+lbl	21	10 - 39
Maldives	2009	7	v023	sample domain	dbl+lbl	21	10 - 39
Maldives	2016	7	v023	stratification used in sample design	dbl+lbl	21	10 - 39
Maldives	2009	8	v024	region	dbl+lbl	6	1 - 6
Maldives	2016	8	v024	region	dbl+lbl	6	1 - 6

From the above we can see that v023 and v024 are of labelled class, while the rest are in numeric class. Therefore, we will check the numeric and labelled variables in different ways. Note that although survey year is a constituent ID variable we have not checked it. It is imperative that survey year would be a 4-digit variable.

Numeric ID variables check

First, let’s find out the required length of the numeric ID variables by checking the maximum values of the constituent ID variable across the Maldives DHS datasets. Here we estimate the summary stats of numeric constituent variables using skim_without_charts().

# Check the summary stats for ID vars using skimr in each mvbr dataset.
# First we estimate the summary stats using skim_without_charts().
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(
    skim_id_num = map(
      mvbr_data,
      function(df) {
        df |> 
          select(v001, v002, v003, bord, v021, v022) |> 
          skim_without_charts() |> 
          as_tibble() |> 
          select(-c(skim_type, n_missing, complete_rate)) |> 
          rename(
            variable = 1,
            mean = 2,
            sd = 3,
            min = 4,
            p25 = 5,
            p50 = 6,
            p75 = 7,
            max = 8
          )
      }
    )
  )
mvbr1_pre_tmp1

Next, we check the summary stats of numeric variables by variable name-wise.

# Now we unnest the nested tibble so that we can compare the variable length 
# across the mvbr datasets.
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(skim_id_num)) |> 
  arrange(variable, svy_year) |> 
  # change the decimal places of selected variables
  mutate(
    mean = sprintf("%.1f", mean),
    sd = sprintf("%.1f", sd),
    p75 = sprintf("%.0f", p75)
  )
# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 4: Summary statistics of the numeric ID variables

ctr_name	svy_year	variable	mean	sd	min	p25	p50	p75	max
Maldives	2009	bord	3.0	2.0	1	1	2	4	14
Maldives	2016	bord	2.3	1.5	1	1	2	3	13
Maldives	2009	v001	141.3	71.9	1	84	143	199	270
Maldives	2016	v001	139.7	71.7	1	79	139	201	266
Maldives	2009	v002	28.4	18.9	1	13	26	41	99
Maldives	2016	v002	13.0	7.4	1	7	13	19	42
Maldives	2009	v003	2.5	2.0	1	2	2	3	23
Maldives	2016	v003	2.4	1.8	1	1	2	3	27
Maldives	2009	v021	141.3	71.9	1	84	143	199	270
Maldives	2016	v021	139.7	71.7	1	79	139	201	266
Maldives	2009	v022	10.1	6.2	1	5	10	15	21
Maldives	2016	v022	27.3	7.9	10	22	27	34	39

Now we find out the required length of the numeric ID variables to be set, so that we can correctly concatenate them to create the ID variables. The required length of the numeric ID variables are given in max_digits column. Note that survey year is also a constituent ID variable of 4-digits.

# Processing the above nested tibble further
mvbr1_pre_tmp3 <- mvbr1_pre_tmp2 |> 
  group_by(variable) |> 
  # find the minimum and maximum values across surveys 
  summarize(
    min_val = min(min),
    max_val = max(max)
  ) |> 
  mutate(
    # calculate the num of digits in the maximum values
    max_digits = nchar(as.character(max_val)),
    # convert char var to factor
    variable = fct(
      variable, 
      levels = c("v001", "v002", "v003", "bord", "v021", "v022")
    )
  ) |> 
  # sort the rows by factor levels 
  arrange(variable) |> 
  # add variable labels and relocate it after variable name.
  bind_cols(vlabel = c("cluster number", "household number", 
                       "respondent's line number", "birth order", 
                       "primary sampling unit", "sample strata for se")) |> 
  relocate(vlabel, .after = 1)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp3 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 5: The maximum length of numeric variables to be set across the mvbr rounds for concatenating the ID variables

variable	vlabel	min_val	max_val	max_digits
v001	cluster number	1	270	3
v002	household number	1	99	2
v003	respondent's line number	1	27	2
bord	birth order	1	14	2
v021	primary sampling unit	1	270	3
v022	sample strata for se	1	39	2

Labelled ID variables check

First we check the labels in sub-national region variable coded as v024 across the mvbr datasets. Let’s create a nested tibble of v024’s value labels.

# Create the data dictionary for v024 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v024 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v024) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

Now we view the value labels of v024 in the table below.

# Now we unnest the tibble and refine the pooled data dictionary 
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v024)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |> 
  # Show the variable name in a col
  mutate(var_name = "v024", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 6: Data dictionary of v024 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v024	1	[1] malé	[1] malé
Maldives	v024	2	[2] north	[2] north region
Maldives	v024	3	[3] north central	[3] north central
Maldives	v024	4	[4] central	[4] central region
Maldives	v024	5	[5] south central	[5] south central
Maldives	v024	6	[6] south	[6] south region

NOTE: The sub-national region var, v024 has different label values in each survey year. It was same for mvbr 1996, 2001 and 2006. After that the label values are different for each survey round.
VERD: In this analysis, we do not use the region var in the ID var.

Secondly, we check the labels in v023 variable that denotes the stratifications used for sampling design. First we create a nested tibble of v023’s value labels.

# Create the data dictionary for v023 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v023 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v023) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

Now we view the value labels of v023 in the table below.

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v023)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v023", .before = 2) 

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 7: Data dictionary of v023 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v023	10	[10] male	[10] malé
Maldives	v023	20	[20] haa alif	[20] north thiladhunmathi (ha)
Maldives	v023	21	[21] haa dhaal	[21] south thiladhunmathi (hdh)
Maldives	v023	22	[22] shaviyani	[22] north miladhunmadulu (sh)
Maldives	v023	23	[23] noonu	[23] south miladhunmadulu (n)
Maldives	v023	24	[24] raa	[24] north maalhosmadulu (r)
Maldives	v023	25	[25] baa	[25] south maalhosmadulu (b)
Maldives	v023	26	[26] lhaviyani	[26] faadhippolhu (lh)
Maldives	v023	27	[27] kaafu	[27] malé atoll (k)
Maldives	v023	28	[28] alif alif	[28] north ari atoll (aa)
Maldives	v023	29	[29] alif dhaal	[29] south ari atoll (adh)
Maldives	v023	30	[30] vaavu	[30] felidhe atoll (v)
Maldives	v023	31	[31] meemu	[31] mulakatholhu (m)
Maldives	v023	32	[32] faafu	[32] north nilandhe atoll (f)
Maldives	v023	33	[33] dhaalu	[33] south nilandhe atoll (dh)
Maldives	v023	34	[34] thaa	[34] kolhumadulu (th)
Maldives	v023	35	[35] lhaamu	[35] hadhdhunmathi (l)
Maldives	v023	36	[36] gaaf alif	[36] north huvadhu atoll (ga)
Maldives	v023	37	[37] gaaf dhaal	[37] south huvadhu atoll (gdh)
Maldives	v023	38	[38] gnaviyani	[38] gnaviyani (gn)
Maldives	v023	39	[39] seenu	[39] addu atoll (s)

NOTE: The labels of v023 are different across the survey rounds.
VERD: Therefore we cannot use v023 in the ID variable preparation.

Checking the Birth History variables before harmonization

Undoubtedly the birth history variables are important for this study objective. Therefore, we need to scrutinize all the birth history variables before using them to prepare harmonized variables for the pooled dataset.

# We check the birth history vars in all mvbr datasets.
# First we create a data dictionary in nested tibble.
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |>
  mutate(lookfor_bhvars = map(
    mvbr_data,
    \(df) {
      df |> 
        select(bidx, matches("^b[0-9]+")) |> 
        lookfor(details = "full") |> 
        select(-c(levels:n_na)) |> 
        # For correctly viewing the range column in data dictionary
        convert_list_columns_to_character()
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_bhvars)) |> 
  arrange(pos)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 8: Data dictionary of birth history variables across the mvbr rounds

ctr_name	svy_year	pos	variable	label	col_type	missing	unique_values	range
Maldives	2009	1	bidx	birth column number	dbl	0	14	1 - 14
Maldives	2016	1	bidx	birth column number	dbl	0	13	1 - 13
Maldives	2009	2	b0	child is twin	dbl+lbl	0	3	0 - 2
Maldives	2016	2	b0	child is twin	dbl+lbl	0	3	0 - 2
Maldives	2009	3	b1	month of birth	dbl	0	12	1 - 12
Maldives	2016	3	b1	month of birth	dbl	0	12	1 - 12
Maldives	2009	4	b2	year of birth	dbl	0	38	1972 - 2009
Maldives	2016	4	b2	year of birth	dbl	0	38	1980 - 2017
Maldives	2009	5	b3	date of birth (cmc)	dbl	0	420	867 - 1316
Maldives	2016	5	b3	date of birth (cmc)	dbl	0	425	972 - 1415
Maldives	2009	6	b4	sex of child	dbl+lbl	0	2	1 - 2
Maldives	2016	6	b4	sex of child	dbl+lbl	0	2	1 - 2
Maldives	2009	7	b5	child is alive	dbl+lbl	0	2	0 - 1
Maldives	2016	7	b5	child is alive	dbl+lbl	0	2	0 - 1
Maldives	2009	8	b6	age at death	dbl+lbl	18847	84	100 - 998
Maldives	2016	8	b6	age at death	dbl+lbl	13452	67	100 - 328
Maldives	2009	9	b7	age at death (months-imputed)	dbl	18836	53	0 - 360
Maldives	2016	9	b7	age at death (months, imputed)	dbl	13452	44	0 - 336
Maldives	2009	10	b8	current age of child	dbl	1300	38	0 - 36
Maldives	2016	10	b8	current age of child	dbl	470	38	0 - 36
Maldives	2009	11	b9	child lives with whom	dbl+lbl	1300	3	0 - 4
Maldives	2016	11	b9	child lives with whom	dbl+lbl	470	3	0 - 4
Maldives	2009	12	b10	completeness of information	dbl+lbl	0	7	1 - 8
Maldives	2016	12	b10	completeness of information	dbl+lbl	0	4	0 - 4
Maldives	2009	13	b11	preceding birth interval	dbl	6149	185	9 - 241
Maldives	2016	13	b11	preceding birth interval (months)	dbl	5451	193	8 - 266
Maldives	2009	14	b12	succeeding birth interval	dbl	6158	185	9 - 241
Maldives	2016	14	b12	succeeding birth interval (months)	dbl	5465	193	8 - 266
Maldives	2009	15	b13	flag for age at death	dbl+lbl	18836	6	0 - 8
Maldives	2016	15	b13	flag for age at death	dbl+lbl	13452	2	0 - 0
Maldives	2009	16	b15	live birth between births	dbl+lbl	6245	3	0 - 1
Maldives	2016	16	b15	live birth between births	dbl+lbl	5408	3	0 - 1
Maldives	2009	17	b16	child's line number in household	dbl+lbl	1300	25	0 - 23
Maldives	2016	17	b16	child's line number in household	dbl+lbl	470	24	0 - 23
Maldives	2016	18	b17	day of birth	dbl	0	31	1 - 31
Maldives	2016	19	b18	century day code of birth (cdc)	dbl	0	7851	29558 - 43060
Maldives	2016	20	b19	current age of child in months (months since birth for dead children)	dbl	0	411	0 - 432
Maldives	2016	21	b20	duration of pregnancy	dbl	10474	8	4 - 10

From the above table we get an overall snapshot of the birth history variables. We see that the variables b1-b13 are common in all the six mvbr datasets. Notably mvbr 2001 and 2006 have some extra variables that are not available in other rounds. Next, we look at the other labelled variables which are common across mvbr in more details. We would like to see if the value labels of the common birth history variables are similar across the mvbr datasets.

b0 - child is twin

We check the value labels of b0 variable that denotes whether the child is twin. First we create a nested tibble of b0’s value labels.

# Create the data dictionary for b0 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_b0 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(b0) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_b0)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "b0", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 9: Data dictionary of b0 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	b0	0	[0] single birth	[0] single birth
Maldives	b0	1	[1] 1st of multiple	[1] 1st of multiple
Maldives	b0	2	[2] 2nd of multiple	[2] 2nd of multiple
Maldives	b0	3	[3] 3rd of multiple	[3] 3rd of multiple
Maldives	b0	4	[4] 4th of multiple	[4] 4th of multiple
Maldives	b0	5	[5] 5th of multiple	[5] 5th of multiple

We can see the value labels of b0 in the above table. We see that the value labels are same across all the mvbr datasets.

b4 - sex of child

We check the value labels of b4 variable which gives the sex of the child. First we create a nested tibble of b4’s value labels.

# Create the data dictionary for b4 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_b4 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(b4) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_b4)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "b4", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 10: Data dictionary of b4 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	b4	1	[1] male	[1] male
Maldives	b4	2	[2] female	[2] female

We can see the value labels of b4 in the above table. The value labels are same across all the mvbr datasets.

b5 - child is alive

We check the value labels of b5 variable which gives the survival status of the child. First we create a nested tibble of b5’s value labels.

# Create the data dictionary for b5 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_b5 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(b5) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_b5)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "b5", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 11: Data dictionary of b5 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	b5	0	[0] no	[0] no
Maldives	b5	1	[1] yes	[1] yes

The above table shows that the value labels of survival status of child are same across all the mvbr datasets.

b6 - age at death

We check the value labels of b6 variable which shows the age at death of children. Note that this variable has many missing values across all mvbr rounds as not all children experienced mortality throughout their lifetime. First we create a nested tibble of b6’s value labels.

# Create the data dictionary for b5 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_b6 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(b6) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_b6)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "b6", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 12: Data dictionary of b6 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	b6	100		[100] died on day of birth
Maldives	b6	101		[101] days: 1
Maldives	b6	199		[199] days: number missing
Maldives	b6	201		[201] months: 1
Maldives	b6	299		[299] months: number missing
Maldives	b6	301		[301] years: 1
Maldives	b6	399		[399] years: number missing
Maldives	b6	997	[997] inconsistent	[997] inconsistent
Maldives	b6	998	[998] don't know	[998] don't know

The above table shows that the value labels of age at death of child are in two groups. First, they are same for mvbr 1996, 2001 and 2006 and and then for mvbr 2011, 2016 and 2022.

b9 - child lives with whom

We check the value labels of b9 variable which gives info on who the child lives with. First we create a nested tibble of b9’s value labels.

# Create the data dictionary for b9 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_b9 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(b9) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_b9)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "b9", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 13: Data dictionary of b9 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	b9	0	[0] respondent	[0] respondent
Maldives	b9	1	[1] father	[1] father
Maldives	b9	2	[2] other relative	[2] other relative
Maldives	b9	3	[3] someone else	[3] someone else
Maldives	b9	4	[4] lives elsewhere	[4] lives elsewhere

We can see in the above table that the value labels of b9 are same across all the mvbr datasets.

b10 - completeness of information

We check the value labels of b10 variable which gives the completeness of birth history information. First we create a nested tibble of b10’s value labels.

# Create the data dictionary for b10 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_b10 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(b10) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_b10)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "b10", .before = 2) 

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |>
  align(align = "left", part = "all") |> 
  autofit()

Table 14: Data dictionary of b10 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	b10	0		[0] month, year and day
Maldives	b10	1	[1] month and year	[1] month and year - information complete
Maldives	b10	2	[2] month and age -y imp	[2] month and age - year imputed
Maldives	b10	3	[3] year and age - m imp	[3] year and age - month imputed
Maldives	b10	4	[4] y & age - y ignored	[4] year and age - year ignored
Maldives	b10	5	[5] year - a, m imp	[5] year - age/month imputed
Maldives	b10	6	[6] age - y, m imp	[6] age - year/month imputed
Maldives	b10	7	[7] month - a, y imp	[7] month - age/year imputed
Maldives	b10	8	[8] none - all imp	[8] none - all imputed

We can see in the above table that the value labels of b10 are same across mvbr 1996, 2001, 2006 and 2011 datasets. Then they are same for mvbr 2016 and 2022.

Checking the Common independent variables before harmonization

Next we start documenting the common independent variables. First we will check the data dictionary of the common independent variables. Then we will check them variable wise.

# We check the common independent vars in all mvbr datasets.
# First we create the data dictionary in nested tibble.
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |>
  mutate(lookfor_comindvars = map(
    mvbr_data,
    \(df) {
      df |> 
        # select the common independent variables
        select(v106, v011, v501, v701, v025, v151, v152, v190) |> 
        lookfor(details = "full") |> 
        select(-c(levels:n_na)) |> 
        # For correctly viewing the range column in data dictionary
        convert_list_columns_to_character()
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary 
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_comindvars)) |> 
  arrange(pos)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 15: Data dictionary of common independent variables across the mvbr rounds

ctr_name	svy_year	pos	variable	label	col_type	missing	unique_values	range
Maldives	2009	1	v106	highest educational level	dbl+lbl	0	5	0 - 8
Maldives	2016	1	v106	highest educational level	dbl+lbl	0	4	0 - 3
Maldives	2009	2	v011	date of birth (cmc)	dbl	0	374	711 - 1093
Maldives	2016	2	v011	date of birth (cmc)	dbl	0	378	797 - 1193
Maldives	2009	3	v501	current marital status	dbl+lbl	0	4	1 - 5
Maldives	2016	3	v501	current marital status	dbl+lbl	0	5	0 - 4
Maldives	2009	4	v701	partner's education level	dbl+lbl	333	6	0 - 8
Maldives	2016	4	v701	husband/partner's education level	dbl+lbl	1192	6	0 - 8
Maldives	2009	5	v025	type of place of residence	dbl+lbl	0	2	1 - 2
Maldives	2016	5	v025	type of place of residence	dbl+lbl	0	2	1 - 2
Maldives	2009	6	v151	sex of household head	dbl+lbl	0	2	1 - 2
Maldives	2016	6	v151	sex of household head	dbl+lbl	0	2	1 - 2
Maldives	2009	7	v152	age of household head	dbl+lbl	4	74	18 - 98
Maldives	2016	7	v152	age of household head	dbl+lbl	0	73	21 - 95
Maldives	2009	8	v190	wealth index	dbl+lbl	0	5	1 - 5
Maldives	2016	8	v190	wealth index combined	dbl+lbl	0	5	1 - 5

From the above table we get an overall snapshot of the common independent variables. We see that majority of the have different number of value labels across the six mvbr datasets. Only v025 and v151 have the same number of value labels across mvbr rounds. Next, we look at the labelled variables among these common variables in more details. We would like to see if the value labels and codes of the common independent variables are similar across the mvbr datasets.

v106 - Mother’s education level

We check the value labels of v106 variable that denotes the highest education level of mother. First we create a nested tibble of v106’s value labels.

# Create the data dictionary for v106 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v106 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v106) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v106)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v106", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 16: Data dictionary of v106 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v106	0	[0] no education	[0] no education
Maldives	v106	1	[1] primary	[1] primary
Maldives	v106	2	[2] secondary	[2] secondary
Maldives	v106	3	[3] higher	[3] higher
Maldives	v106	8	[8] unknown - certificate

We can see the value labels of v106 are mostly similar except for mvbr 1996 and 2011 datasets.

v011 - Date of birth (in CMC)

The v011 variable, which has the dob of mothers in cmc, is a numeric variable. Let’s check the range of these values in further details such as checking for outliers. First let’s create a nested tibble of the summary statistics of v011 variable.

# Create the summary statistics for v011 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(skim_v011 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v011) |> 
        skim_without_charts() |> 
        as_tibble() |> 
        select(-c(skim_type, complete_rate)) |> 
        rename(
          variable = 1,
          n_miss = 2,
          mean = 3,
          sd = 4,
          min = 5,
          p25 = 6,
          p50 = 7,
          p75 = 8,
          max = 9
        )
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(skim_v011)) |> 
  # Make variable values have one decimal point 
  mutate(
    mean = sprintf("%.1f", mean),
    sd = sprintf("%.1f", sd)
  )

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 17: Data dictionary of v011 across the mvbr rounds

ctr_name	svy_year	variable	n_miss	mean	sd	min	p25	p50	p75	max
Maldives	2009	v011	0	852.4	87.5	711	781	842	917	1093
Maldives	2016	v011	0	951.2	87.6	797	873	948	1022	1193

v501 - Mother’s marital status

We check the value labels of v501 variable which gives the current marital status of mother. First we create a nested tibble of v501’s value labels.

# Create the data dictionary for v501 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v501 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v501) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v501)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v501", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 18: Data dictionary of v501 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v501	0	[0] never married	[0] never in union
Maldives	v501	1	[1] married	[1] married
Maldives	v501	2	[2] living together	[2] living with partner
Maldives	v501	3	[3] widowed	[3] widowed
Maldives	v501	4	[4] divorced	[4] divorced
Maldives	v501	5	[5] not living together	[5] no longer living together/separated

All the mvbr rounds have 5 value labels. The mvbr 1996, 2001 and 2006 rounds have a set of similar value label texts. Then mvbr 2011, 2016 and 2022 have another set of similar value labels.

v701 - Husband/Partner’s education level

We check the value labels of v701 variable which gives the current marital status of mother. First we create a nested tibble of v701’s value labels.

# Create the data dictionary for v701 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v701 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v701) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v701)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v701", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 19: Data dictionary of v701 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v701	0	[0] no education	[0] no education
Maldives	v701	1	[1] primary	[1] primary
Maldives	v701	2	[2] secondary	[2] secondary
Maldives	v701	3	[3] higher	[3] higher
Maldives	v701	8	[8] don't know	[8] don't know

All the mvbr rounds have 5 value labels. The mvbr 1996, 2001 and 2006 rounds and mvbr 2011, 2016 and 2022 have a similar set of value labels with a difference in wording among them.

v025 - Type of place of residence

We check the value labels of v025 variable which shows if a household belongs to rural or urban psu. First we create a nested tibble of v025’s value labels.

# Create the data dictionary for v025 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v025 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v025) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(c(ctr_name, svy_year, lookfor_v025)) |> 
  unnest(cols = c(lookfor_v025)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v025", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 20: Data dictionary of v025 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v025	1	[1] urban	[1] urban
Maldives	v025	2	[2] rural	[2] rural

The values labels and codes for v025 are same across all the mvbr rounds.

v151 - Sex of household head

We check the value labels of v151 variable which gives the gender of the household head. First we create a nested tibble of v151’s value labels, then pivot wide and compare.

# Create the data dictionary for v151 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_v151 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v151) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v151)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v151", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 21: Data dictionary of v151 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v151	1	[1] male	[1] male
Maldives	v151	2	[2] female	[2] female

The values labels and codes for v151 are same across all the mvbr rounds.

v152 - Age of household head

Interestingly, we see v152 (a continuous variable) has value labels for all rounds except mvbr 1996. Therefore, we check the value labels of v152 for those rounds. First we create a nested tibble of v152’s value labels.

# Create the data dictionary for v152 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  filter(svy_year != 1996) |> 
  mutate(lookfor_v152 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v152) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v152)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v152", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 22: Data dictionary of v152 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v152	97	[97] 97+	[97] 97+
Maldives	v152	98	[98] dk	[98] don't know

We can see that the value labels of v152 are mostly for missing values. However, since v152 has no missing values across the mvbr rounds, we need not be concerned about them.

v190 - Wealth quintile of household

We check the value labels of v190 variable which gives the gender of the household head. First we create a nested tibble of v190’s value labels, then pivot wide and compare.

# Create the data dictionary for v190 in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  filter(svy_year != 1996) |> 
  mutate(lookfor_v190 = map(
    mvbr_data,
    \(df) {
      df |> 
        select(v190) |> 
        look_for() |> 
        lookfor_to_long_format() |> 
        select(value_labels)
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_v190)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvbr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "v190", .before = 2)

# Convert the tibble to flextable for easy viewing
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 23: Data dictionary of v190 across the mvbr rounds

ctr_name	var_name	label_num	mvbr_2009	mvbr_2016
Maldives	v190	1	[1] poorest	[1] poorest
Maldives	v190	2	[2] poorer	[2] poorer
Maldives	v190	3	[3] middle	[3] middle
Maldives	v190	4	[4] richer	[4] richer
Maldives	v190	5	[5] richest	[5] richest

The values labels and codes for v190 are same across all the mvbr rounds.

Correcting year-related variables

The year-related variables might have different formatting in each survey. Therefore, we need to check and harmonize them before appending the datasets.

# First we create the data dictionary of year-related vars in nested tibble
mvbr1_pre_tmp1 <- mvbr1_pre_tmp0 |> 
  mutate(lookfor_year = map(
    mvbr_data,
    \(df) {
      df |> 
        select(c(b2, v007, v010)) |> 
        look_for(details = "full") |> 
        # For correctly viewing the range column in data dictionary
        convert_list_columns_to_character() |> 
        select(-c(levels:n_na))
    }
  ))
mvbr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary 
mvbr1_pre_tmp2 <- mvbr1_pre_tmp1 |> 
  select(-c(unf, mvbr_data, n_births)) |> 
  unnest(cols = c(lookfor_year)) |> 
  arrange(pos)
# Convert and view the tibble as flextable
mvbr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 24: Data dictionary of year-related variables across the mvbr rounds

ctr_name	svy_year	pos	variable	label	col_type	unique_values	range
Maldives	2009	1	b2	year of birth	dbl	38	1972 - 2009
Maldives	2016	1	b2	year of birth	dbl	38	1980 - 2017
Maldives	2009	2	v007	year of interview	dbl	1	2009 - 2009
Maldives	2016	2	v007	year of interview	dbl	2	2016 - 2017
Maldives	2009	3	v010	respondent's year of birth	dbl	33	1959 - 1991
Maldives	2016	3	v010	respondent's year of birth	dbl	34	1966 - 1999

We can say that the year variables have correct values, especially by looking at the year of interview variable. Therefore, we use the year variables directly for analysis.

Maldives PR dataset use for family structure variables creation

Checking the ID variables before harmonization

Here we check the formatting of the constituent variables with which we will prepare the ID variables for the pooled Maldives person recode (pr) dataset. We will use the following constituent variables for creating the ID variables for the pooled dataset:

# We check the var type of ID vars in all mvpr datasets.
# First we create a data dictionary of the mvpr datasets in nested tibble.
mvpr1_pre_tmp1 <- mvpr1_pre_tmp0 |>
  mutate(lookfor_idvars = map(mvpr_data, \(df) {
    df |> 
      select(hv001, hv002, hvidx) |> 
      lookfor(details = "full") |> 
      select(-c(levels:n_na)) |> 
      # For correctly viewing the range column in data dictionary
      convert_list_columns_to_character()
  }))
mvpr1_pre_tmp1

# Now we unnest the tibble and output the pooled data dictionary 
mvpr1_pre_tmp2 <- mvpr1_pre_tmp1 |> 
  select(c(ctr_name, svy_year, lookfor_idvars)) |> 
  unnest(cols = c(lookfor_idvars)) |> 
  arrange(pos)

# Convert and view the tibble as flextable
mvpr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 25: Data dictionary of variables to be used for ID creation across the mvpr rounds

ctr_name	svy_year	pos	variable	label	col_type	unique_values	range
Maldives	2009	1	hv001	cluster number	dbl	270	1 - 270
Maldives	2016	1	hv001	cluster number	dbl	266	1 - 266
Maldives	2009	2	hv002	household number	dbl	99	1 - 99
Maldives	2016	2	hv002	household number	dbl	43	1 - 43
Maldives	2009	3	hvidx	line number	dbl	32	1 - 32
Maldives	2016	3	hvidx	line number	dbl	30	1 - 30

From the above table we can see that all the three constituent ID variables are of numeric class with no missing values. These variables can directly be used for preparing the ID variables after finding the maximum length of their largest value. Note that survey year is also a constituent ID variable of 4-digits and we need not check it.

# We thought to process the above nested tibble further by decomposing the 
# "range" col into min and max values using separate_wider_regex().
# However, we hit a roadblock as pattern did not identify the max values in 
# some mvpr rounds correctly
mvpr1_pre_tmp3 <- mvpr1_pre_tmp0 |> 
  # Generate the summary stats for id vars
  mutate(skim_idvars = map(mvpr_data, \(df) {
    df |> 
      select(hv001, hv002, hvidx) |> 
      skim_without_charts()
  })) |> 
  # Pool the summary stats for all mvpr rounds
  select(c(ctr_name, svy_year, skim_idvars)) |> 
  unnest(cols = c(skim_idvars)) |> 
  arrange(skim_variable, svy_year) |> 
  # Group and generate the max and min values for each variable
  group_by(variable = skim_variable) |> 
  summarize(
    min_val = min(numeric.p0),
    max_val = max(numeric.p100)
  ) |> 
  # calculate the num of digits in the maximum values
  mutate(
    max_digits = nchar(as.character(max_val))
  ) |>
  # add variable labels and relocate it after variable name
  bind_cols(vlabel = c("cluster number", "household number", "Persons line number")) |>
  relocate(vlabel, .after = 1)

# Convert the tibble to flextable for easy viewing
mvpr1_pre_tmp3 |>
  qflextable() |>
  align(align = "left", part = "all") |>
  autofit()

Table 26: The maximum length of constituent ID variables to be set across the mvpr rounds

variable	vlabel	min_val	max_val	max_digits
hv001	cluster number	1	270	3
hv002	household number	1	99	2
hvidx	Persons line number	1	32	2

The above table gives the required length of the constituent ID variables to be set, so that we can correctly concatenate them to create the ID variables. The required length of the ID variables are given in max_digits column. Note that survey year is also a constituent ID variable of 4-digits.

Checking Family structure variables before harmonization

Here we check the family structure related variables before harmonizing them. The variable names were collected by manually checking the full data dictionaries. Here we will check the data dictionary of these hh-level variables and focus on the variable types.

# We check the family structure vars in all mvpr datasets.
# First we create the data dictionary in nested tibble.
mvpr1_pre_tmp1 <- mvpr1_pre_tmp0 |>
  mutate(lookfor_famstrvars = map(mvpr_data, \(df) {
    df |> 
      # select the common independent variables
      select(c(hv101, hv102, hv103, hv104, hv105)) |> 
      lookfor(details = "full") |> 
      select(-c(levels:n_na)) |> 
      # For correctly viewing the range column in data dictionary
      convert_list_columns_to_character()
  }))
mvpr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary 
mvpr1_pre_tmp2 <- mvpr1_pre_tmp1 |> 
  select(c(svy_year, lookfor_famstrvars)) |> 
  unnest(cols = c(lookfor_famstrvars)) |> 
  arrange(pos, svy_year)

# Convert the tibble to flextable for easy viewing
mvpr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 27: Data dictionary of family structure vars across the mvpr rounds

svy_year	pos	variable	label	col_type	missing	unique_values	range
2009	1	hv101	relationship to head	dbl+lbl	6	15	1 - 98
2016	1	hv101	relationship to head	dbl+lbl	0	12	1 - 98
2009	2	hv102	usual resident	dbl+lbl	18	3	0 - 1
2016	2	hv102	usual resident	dbl+lbl	0	2	0 - 1
2009	3	hv103	slept last night	dbl+lbl	22	3	0 - 1
2016	3	hv103	slept last night	dbl+lbl	0	2	0 - 1
2009	4	hv104	sex of household member	dbl+lbl	2	3	1 - 2
2016	4	hv104	sex of household member	dbl+lbl	0	2	1 - 2
2009	5	hv105	age of household members	dbl+lbl	538	96	0 - 98
2016	5	hv105	age of household members	dbl+lbl	0	96	0 - 95

The above table gives an overall snapshot of the family structure related variables. Interestingly, all the variables including age of hh members (a continuous var) are of labelled class. The relation to head and de facto resident variables have few missing values in mvpr 1996. Note that, the three variables of interest hv101-hv102, two variables hv101 and hv103 have different number of value labels across the mvpr rounds. Next, we compare the value labels of the individual variables across the mvpr datasets.

hv101 - Relationship to head

Next, we check the value labels of the relationship to the household head variable. First we create a nested tibble of the value labels.

# Create the data dictionary in nested tibble
mvpr1_pre_tmp1 <- mvpr1_pre_tmp0 |> 
  mutate(lookfor_hv101 = map(mvpr_data, \(df) {
    df |> 
      select(hv101) |> 
      look_for() |> 
      lookfor_to_long_format() |> 
      select(value_labels)
  }))
mvpr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvpr1_pre_tmp2 <- mvpr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(c(ctr_name, svy_year, lookfor_hv101)) |> 
  unnest(cols = c(lookfor_hv101)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvpr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "hv101", .before = 2)

# Convert the tibble to flextable for easy viewing
mvpr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 28: Data dictionary of relationship to head variable across the mvpr rounds

ctr_name	var_name	label_num	mvpr_2009	mvpr_2016
Maldives	hv101	1	[1] head	[1] head
Maldives	hv101	2	[2] wife or husband	[2] wife or husband
Maldives	hv101	3	[3] son/daughter	[3] son/daughter
Maldives	hv101	4	[4] son/daughter-in-law	[4] son/daughter-in-law
Maldives	hv101	5	[5] grandchild	[5] grandchild
Maldives	hv101	6	[6] parent	[6] parent
Maldives	hv101	7	[7] parent-in-law	[7] parent-in-law
Maldives	hv101	8	[8] brother/sister	[8] brother/sister
Maldives	hv101	9	[9] co-spouse	[9] co-spouse
Maldives	hv101	10	[10] other relative	[10] other relative
Maldives	hv101	11	[11] adopted/foster child	[11] adopted/foster child
Maldives	hv101	12	[12] not related	[12] not related
Maldives	hv101	13	[13] niece/nephew by blood	[13] niece/nephew by blood
Maldives	hv101	14	[14] niece/nephew by marriage	[14] niece/nephew by marriage
Maldives	hv101	98	[98] dk	[98] don't know

The above table shows that the value label texts vary across the mvpr rounds. To harmonize the relationship to head variable we can use the following value labels -

1 head
2 spouse
3 child
4 child-in-law
5 grandchild
6 parent
7 parent-in-law
8 sibling
9 others

Here, we merge the “spouse” and “co-spouse” categories into “spouse” category, and the “son/daughter” and “adopted/foster child” categories into “child” category.

hv102 - de jure/usual resident

Next, we check the value labels of the de jure resident variable. This means if a household member is an usual resident of the household. First we create a nested tibble of the value labels.

# Create the data dictionary in nested tibble
mvpr1_pre_tmp1 <- mvpr1_pre_tmp0 |> 
  mutate(lookfor_hv102 = map(mvpr_data, \(df) {
    df |> 
      select(hv102) |> 
      look_for() |> 
      lookfor_to_long_format() |> 
      select(value_labels)
  }))
mvpr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvpr1_pre_tmp2 <- mvpr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(c(ctr_name, svy_year, lookfor_hv102)) |> 
  unnest(cols = c(lookfor_hv102)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvpr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "hv102", .before = 2)

# Convert the tibble to flextable for easy viewing
mvpr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 29: Data dictionary of the De jure resident variable across the mvpr rounds

ctr_name	var_name	label_num	mvpr_2009	mvpr_2016
Maldives	hv102	0	[0] no	[0] no
Maldives	hv102	1	[1] yes	[1] yes

The above table shows that hv102 has the same value label texts and codes across the mvpr rounds. Therefore, we can use this variable directly after converting to factor type.

hv103 - de facto resident

Next, we check the value labels of the de facto resident variable. In DHS this means if a household member slept last night in the household. First we create a nested tibble of the value labels.

# Create the data dictionary in nested tibble
mvpr1_pre_tmp1 <- mvpr1_pre_tmp0 |> 
  mutate(lookfor_hv103 = map(mvpr_data, \(df) {
    df |> 
      select(hv103) |> 
      look_for() |> 
      lookfor_to_long_format() |> 
      select(value_labels)
  }))
mvpr1_pre_tmp1

# Now we unnest the tibble and refine the pooled data dictionary
mvpr1_pre_tmp2 <- mvpr1_pre_tmp1 |> 
  # First we select the required cols and unnest()
  select(c(ctr_name, svy_year, lookfor_hv103)) |> 
  unnest(cols = c(lookfor_hv103)) |> 
  # Next we make the num of value labels same across each round
  mutate(label_num = parse_number(value_labels)) |> 
  complete(ctr_name, svy_year, label_num) |>
  # Next we create col of value labels for each survey round
  pivot_wider(
    names_from = svy_year, 
    values_from = value_labels,
    names_prefix = "mvpr_"
  ) |>
  # Show the variable name in a col
  mutate(var_name = "hv103", .before = 2)

# Convert the tibble to flextable for easy viewing
mvpr1_pre_tmp2 |> 
  qflextable() |> 
  align(align = "left", part = "all") |> 
  autofit()

Table 30: Data dictionary of the De facto resident variable across the mvpr rounds

ctr_name	var_name	label_num	mvpr_2009	mvpr_2016
Maldives	hv103	0	[0] no	[0] no
Maldives	hv103	1	[1] yes	[1] yes

The above table shows that hv103 has the same value label texts and codes across the mvpr rounds. Therefore, we can use this variable directly after converting to factor type.

START FROM HERE

TASK:

Handling multiple births in death scarring vars may not be necessary.
Preceding birth interval construction has changed with DHS-7. We could re-construct it.

TO BE CONTINUED …

Getting started

Document introduction

Data import

Maldives BR dataset use for variable creation

Checking the Women’s weight variable before harmonization

Checking the ID variables before harmonization

Numeric ID variables check

Labelled ID variables check

Checking the Birth History variables before harmonization

b0 - child is twin

b4 - sex of child

b5 - child is alive

b6 - age at death

b9 - child lives with whom

b10 - completeness of information

Checking the Common independent variables before harmonization

v106 - Mother’s education level

v011 - Date of birth (in CMC)

v501 - Mother’s marital status

v701 - Husband/Partner’s education level

v025 - Type of place of residence

v151 - Sex of household head

v152 - Age of household head

v190 - Wealth quintile of household

Checking the Social group variables before harmonization

Correcting year-related variables

Maldives PR dataset use for family structure variables creation

Checking the ID variables before harmonization

Checking Family structure variables before harmonization

hv101 - Relationship to head

hv102 - de jure/usual resident

hv103 - de facto resident

START FROM HERE