retrospective-harmonization | Reprex

Where Are People More Likely To Treat Climate Change as the Most Serious Global Problem?

Sat, 06 Mar 2021 00:00:00 +0000

library(regions)
library(lubridate)
library(dplyr)
if ( dir.exists('data-raw') ) {
data_raw_dir <- "data-raw"
} else {
data_raw_dir <- file.path("..", "..", "data-raw")
}

The first results of our longitudinal table were difficult to map, because the surveys used an obsolete regional coding. We will adjust the wrong coding, when possible, and join the data with the European Environment Agency’s (EEA) Air Quality e-Reporting (AQ e-Reporting) data on environmental pollution. We recoded the annual level for every available reporting stations [not shown here] and all values are in μg/m3. The period under observation is 2014-2016. Data file: https://www.eea.europa.eu/data-and-maps/data/aqereporting-8 (European Environment Agency 2021).

Recoding the Regions

Recoding means that the boundaries are unchanged, but the country changed the names and codes of regions because there were other boundary changes which did not affect our observation unit. We explain the problem and the solution in greater detail in our tutorial that aggregates the data on regional levels.

panel <- readRDS((file.path(data_raw_dir, "climate-panel.rds")))
climate_data_geocode <- panel %>%
mutate ( year: lubridate::year(date_of_interview)) %>%
recode_nuts()

Let’s join the air pollution data and join it by corrected geocodes:

load(file.path("data", "air_pollutants.rda")) ## good practice to use system-independent file.path
climate_awareness_air <- climate_data_geocode %>%
rename ( region_nuts_codes : .data$code_2016) %>%
left_join ( air_pollutants, by: "region_nuts_codes" ) %>%
select ( -all_of(c("w1", "wex", "date_of_interview",
"typology", "typology_change", "geo", "region"))) %>%
mutate (
# remove special labels and create NA_numeric_
age_education: retroharmonize::as_numeric(age_education)) %>%
mutate_if ( is.character, as.factor) %>%
mutate (
# we only have responses from 4 years, and this should be treated as a categorical variable
year: as.factor(year)
) %>%
filter ( complete.cases(.) )

The climate_awareness_air data frame contains the answers of 75086 individual respondents. 17.07% thought that climate change was the most serious world problem and 33.6% mentioned climate change as one of the three most important global problems.

summary ( climate_awareness_air )
## rowid serious_world_problems_first
## ZA5877_v2-0-0_1 : 1 Min. :0.0000
## ZA5877_v2-0-0_10 : 1 1st Qu.:0.0000
## ZA5877_v2-0-0_100 : 1 Median :0.0000
## ZA5877_v2-0-0_1000 : 1 Mean :0.1707
## ZA5877_v2-0-0_10000: 1 3rd Qu.:0.0000
## ZA5877_v2-0-0_10001: 1 Max. :1.0000
## (Other) :75080
## serious_world_problems_climate_change isocntry
## Min. :0.000 BE : 3028
## 1st Qu.:0.000 CZ : 3023
## Median :0.000 NL : 3019
## Mean :0.336 SK : 3000
## 3rd Qu.:1.000 SE : 2980
## Max. :1.000 DE-W : 2978
## (Other):57058
## marital_status age_education
## (Re-)Married: without children :13242 18 :15485
## (Re-)Married: children this marriage :12696 19 : 7728
## Single: without children : 7650 16 : 5840
## (Re-)Married: w children of this marriage: 6520 still studying: 5098
## (Re-)Married: living without children : 6225 17 : 5092
## Single: living without children : 4102 15 : 4528
## (Other) :24651 (Other) :31315
## age_exact occupation_of_respondent
## Min. :15.0 Retired, unable to work :22911
## 1st Qu.:36.0 Skilled manual worker : 6774
## Median :51.0 Employed position, at desk : 6716
## Mean :50.1 Employed position, service job: 5624
## 3rd Qu.:65.0 Middle management, etc. : 5252
## Max. :99.0 Student : 5098
## (Other) :22711
## occupation_of_respondent_recoded
## Employed (10-18 in d15a) :32763
## Not working (1-4 in d15a) :37125
## Self-employed (5-9 in d15a): 5198
##
##
##
##
## respondent_occupation_scale_c_14
## Retired (4 in d15a) :22911
## Manual workers (15 to 18 in d15a) :15269
## Other white collars (13 or 14 in d15a): 9203
## Managers (10 to 12 in d15a) : 8291
## Self-employed (5 to 9 in d15a) : 5198
## Students (2 in d15a) : 5098
## (Other) : 9116
## type_of_community is_student no_education
## DK : 34 Min. :0.0000 Min. :0.000000
## Large town :20939 1st Qu.:0.0000 1st Qu.:0.000000
## Rural area or village :24686 Median :0.0000 Median :0.000000
## Small or middle sized town: 9850 Mean :0.0679 Mean :0.008151
## Small/middle town :19577 3rd Qu.:0.0000 3rd Qu.:0.000000
## Max. :1.0000 Max. :1.000000
##
## education year region_nuts_codes country_code
## Min. :14.00 2013:25103 LU : 1432 DE : 4531
## 1st Qu.:17.00 2015: 0 MT : 1398 GB : 3538
## Median :18.00 2017:25053 CY : 1192 BE : 3028
## Mean :19.61 2019:24930 SK02 : 1053 CZ : 3023
## 3rd Qu.:22.00 EL30 : 974 NL : 3019
## Max. :30.00 EE : 973 SK : 3000
## (Other):68064 (Other):54947
## pm2_5 pm10 o3 BaP
## Min. : 2.109 Min. : 5.883 Min. : 66.37 Min. :0.0102
## 1st Qu.: 9.374 1st Qu.: 28.326 1st Qu.: 90.89 1st Qu.:0.1779
## Median :11.866 Median : 33.673 Median :102.81 Median :0.4105
## Mean :12.954 Mean : 38.637 Mean :101.49 Mean :0.8759
## 3rd Qu.:15.890 3rd Qu.: 49.488 3rd Qu.:110.73 3rd Qu.:1.0692
## Max. :41.293 Max. :123.239 Max. :141.04 Max. :7.8050
##
## so2 ap_pc1 ap_pc2 ap_pc3
## Min. : 0.0000 Min. :-4.6669 Min. :-2.21851 Min. :-2.1007
## 1st Qu.: 0.0000 1st Qu.:-0.4624 1st Qu.:-0.49130 1st Qu.:-0.5695
## Median : 0.0000 Median : 0.4263 Median : 0.02902 Median :-0.1113
## Mean : 0.1032 Mean : 0.1031 Mean : 0.04166 Mean :-0.1746
## 3rd Qu.: 0.0000 3rd Qu.: 0.9748 3rd Qu.: 0.57416 3rd Qu.: 0.3309
## Max. :42.5325 Max. : 2.0344 Max. : 3.25841 Max. : 4.1615
##
## ap_pc4 ap_pc5
## Min. :-1.7387 Min. :-2.75079
## 1st Qu.:-0.1669 1st Qu.:-0.18748
## Median : 0.0371 Median : 0.01811
## Mean : 0.1154 Mean : 0.06797
## 3rd Qu.: 0.3050 3rd Qu.: 0.34937
## Max. : 3.2476 Max. : 1.42816
##

Let’s see a simple CART tree! We remove the regional codes, because there are very serious differences among regional climate awareness. These differences, together with education level, and the year we are talking about, are the most important predictors of thinking about climate change as the most important global problem in Europe.

# Classification Tree with rpart
library(rpart)
# grow tree
fit <- rpart(as.factor(serious_world_problems_first) ~ .,
method="class", data=climate_awareness_air %>%
select ( - all_of(c("rowid", "region_nuts_codes"))),
control: rpart.control(cp: 0.005))
printcp(fit) # display the results
##
## Classification tree:
## rpart(formula: as.factor(serious_world_problems_first) ~ .,
## data: climate_awareness_air %>% select(-all_of(c("rowid",
## "region_nuts_codes"))), method: "class", control: rpart.control(cp: 0.005))
##
## Variables actually used in tree construction:
## [1] age_education isocntry
## [3] serious_world_problems_climate_change year
##
## Root node error: 12817/75086: 0.1707
##
## n= 75086
##
## CP nsplit rel error xerror xstd
## 1 0.0240566 0 1.00000 1.00000 0.0080438
## 2 0.0082703 3 0.92783 0.92783 0.0078055
## 3 0.0050000 5 0.91129 0.91425 0.0077588
plotcp(fit) # visualize cross-validation results

summary(fit) # detailed summary of splits
## Call:
## rpart(formula: as.factor(serious_world_problems_first) ~ .,
## data: climate_awareness_air %>% select(-all_of(c("rowid",
## "region_nuts_codes"))), method: "class", control: rpart.control(cp: 0.005))
## n= 75086
##
## CP nsplit rel error xerror xstd
## 1 0.024056592 0 1.0000000 1.0000000 0.008043837
## 2 0.008270266 3 0.9278302 0.9278302 0.007805478
## 3 0.005000000 5 0.9112897 0.9142545 0.007758824
##
## Variable importance
## serious_world_problems_climate_change isocntry
## 31 26
## country_code BaP
## 20 8
## pm2_5 ap_pc1
## 4 3
## age_education pm10
## 2 2
## education ap_pc2
## 2 1
## year
## 1
##
## Node number 1: 75086 observations, complexity param=0.02405659
## predicted class=0 expected loss=0.1706976 P(node): 1
## class counts: 62269 12817
## probabilities: 0.829 0.171
## left son=2 (25229 obs) right son=3 (49857 obs)
## Primary splits:
## serious_world_problems_climate_change < 0.5 to the right, improve=2214.2040, (0 missing)
## isocntry splits as RRLLLRRRLLRLRLLLLLLLLLLRRLLLRLL, improve= 728.0160, (0 missing)
## country_code splits as RRLLLRRLLRLLLLLLLLLLRRLLLRLL, improve= 673.3656, (0 missing)
## BaP < 0.4300347 to the right, improve= 310.6229, (0 missing)
## pm2_5 < 13.38264 to the right, improve= 296.4013, (0 missing)
## Surrogate splits:
## age_education splits as ----RRRRRR-RRRRRRRRRR-RRRRRRRRRR-RRRRRRRRRR-RRRRRRRRRR-RRRRRL-RRR-RRRRRRRRR--RRRLLR--R-R, agree=0.664, adj=0, (0 split)
## pm10 < 7.491315 to the left, agree=0.664, adj=0, (0 split)
##
## Node number 2: 25229 observations
## predicted class=0 expected loss=0 P(node): 0.3360014
## class counts: 25229 0
## probabilities: 1.000 0.000
##
## Node number 3: 49857 observations, complexity param=0.02405659
## predicted class=0 expected loss=0.2570752 P(node): 0.6639986
## class counts: 37040 12817
## probabilities: 0.743 0.257
## left son=6 (34631 obs) right son=7 (15226 obs)
## Primary splits:
## isocntry splits as RRLLLRRRLLRLRLLLLLLLLLLRRLLLRLL, improve=1454.9460, (0 missing)
## country_code splits as RRLLLRRLLRLLLLLLLLLLRRLLLRLL, improve=1359.7210, (0 missing)
## BaP < 0.4300347 to the right, improve= 629.8844, (0 missing)
## pm2_5 < 13.38264 to the right, improve= 555.7484, (0 missing)
## ap_pc1 < -0.005459537 to the left, improve= 533.3579, (0 missing)
## Surrogate splits:
## country_code splits as RRLLLRRLLRLLLLLLLLLLRRLLLRLL, agree=0.987, adj=0.957, (0 split)
## BaP < 0.1749425 to the right, agree=0.775, adj=0.264, (0 split)
## pm2_5 < 5.206993 to the right, agree=0.737, adj=0.140, (0 split)
## ap_pc1 < 1.405527 to the left, agree=0.733, adj=0.126, (0 split)
## pm10 < 25.31211 to the right, agree=0.718, adj=0.076, (0 split)
##
## Node number 6: 34631 observations
## predicted class=0 expected loss=0.1769802 P(node): 0.4612178
## class counts: 28502 6129
## probabilities: 0.823 0.177
##
## Node number 7: 15226 observations, complexity param=0.02405659
## predicted class=0 expected loss=0.4392487 P(node): 0.2027808
## class counts: 8538 6688
## probabilities: 0.561 0.439
## left son=14 (11607 obs) right son=15 (3619 obs)
## Primary splits:
## isocntry splits as LL---LLR--L-L----------LL---R--, improve=337.5462, (0 missing)
## country_code splits as LL---LR--L-L--------LL---R--, improve=337.5462, (0 missing)
## age_education splits as ----LLLLLL-LLLRRRRRRR-RRRRRRRRRL-RRRRRRLLRR-RRRRLLRLRL-RRLRRR-RRR-LLLLRRR-----LR-----L-R, improve=294.0807, (0 missing)
## education < 22.5 to the left, improve=262.3747, (0 missing)
## BaP < 0.053328 to the right, improve=232.7043, (0 missing)
## Surrogate splits:
## BaP < 0.053328 to the right, agree=0.878, adj=0.485, (0 split)
## pm2_5 < 4.810361 to the right, agree=0.827, adj=0.271, (0 split)
## ap_pc2 < 0.8746175 to the left, agree=0.792, adj=0.124, (0 split)
## so2 < 0.3302972 to the left, agree=0.781, adj=0.078, (0 split)
## age_education splits as ----LLLLLL-LLLLLLLRLR-LRRLRRRRRR-RRRRLLLLLR-LRLRLLRRLL-LLRLLR-LLR-RRLLLLL-----RR-----R-L, agree=0.779, adj=0.071, (0 split)
##
## Node number 14: 11607 observations, complexity param=0.008270266
## predicted class=0 expected loss=0.3804601 P(node): 0.1545827
## class counts: 7191 4416
## probabilities: 0.620 0.380
## left son=28 (7462 obs) right son=29 (4145 obs)
## Primary splits:
## age_education splits as ----LLLLLL-LRRRRRRRRR-RRLRRLRRLL-RRRRLRLLRR-RLRLLLRLRL-RR-RR--RRL-L-LLRRR------------L-R, improve=123.71070, (0 missing)
## year splits as R-LR, improve=107.79460, (0 missing)
## education < 20.5 to the left, improve= 90.28724, (0 missing)
## occupation_of_respondent splits as LRRLRRRRRLRLLLRLLL, improve= 84.62865, (0 missing)
## respondent_occupation_scale_c_14 splits as LRLLLRRL, improve= 68.88653, (0 missing)
## Surrogate splits:
## education < 20.5 to the left, agree=0.950, adj=0.861, (0 split)
## occupation_of_respondent splits as LLLLRLLRRLRLLLRLLL, agree=0.738, adj=0.267, (0 split)
## respondent_occupation_scale_c_14 splits as LRLLLLRL, agree=0.733, adj=0.251, (0 split)
## is_student < 0.5 to the left, agree=0.709, adj=0.186, (0 split)
## age_exact < 23.5 to the right, agree=0.676, adj=0.094, (0 split)
##
## Node number 15: 3619 observations
## predicted class=1 expected loss=0.3722023 P(node): 0.04819807
## class counts: 1347 2272
## probabilities: 0.372 0.628
##
## Node number 28: 7462 observations
## predicted class=0 expected loss=0.326052 P(node): 0.09937938
## class counts: 5029 2433
## probabilities: 0.674 0.326
##
## Node number 29: 4145 observations, complexity param=0.008270266
## predicted class=0 expected loss=0.4784077 P(node): 0.05520337
## class counts: 2162 1983
## probabilities: 0.522 0.478
## left son=58 (2573 obs) right son=59 (1572 obs)
## Primary splits:
## year splits as L-LR, improve=40.13885, (0 missing)
## occupation_of_respondent splits as LRLLRRRRRLRLLLRLLL, improve=18.33254, (0 missing)
## marital_status splits as LRRRLRRRLRRLRLLRRRRRRLRLRLLRR, improve=17.86888, (0 missing)
## type_of_community splits as LRLRL, improve=17.55254, (0 missing)
## age_education splits as ------------LLRRRRRRR-RR-RL-RR---LRRR-R--LR-R-R---R-R--RR-RR--RR------RRR--------------R, improve=14.66121, (0 missing)
## Surrogate splits:
## type_of_community splits as LLLRL, agree=0.777, adj=0.412, (0 split)
## marital_status splits as RRLLLLLRLLLLLLLRRRLLLLLLRLRLL, agree=0.680, adj=0.155, (0 split)
## isocntry splits as LL---LL---L-R----------LL------, agree=0.669, adj=0.127, (0 split)
## country_code splits as LL---L---L-R--------LL------, agree=0.669, adj=0.127, (0 split)
## o3 < 83.06345 to the right, agree=0.650, adj=0.076, (0 split)
##
## Node number 58: 2573 observations
## predicted class=0 expected loss=0.4240187 P(node): 0.03426737
## class counts: 1482 1091
## probabilities: 0.576 0.424
##
## Node number 59: 1572 observations
## predicted class=1 expected loss=0.43257 P(node): 0.02093599
## class counts: 680 892
## probabilities: 0.433 0.567
# plot tree
plot(fit, uniform=TRUE,
main="Classification Tree: Climate Change Is The Most Serious Threat")
text(fit, use.n=TRUE, all=TRUE, cex=.8)
## Warning in labels.rpart(x, minlength: minlength): more than 52 levels in a
## predicting factor, truncated for printout

saveRDS ( climate_awareness_air , file.path(tempdir(), "climate_panel_recoded.rds"), version: 2)
# not evaluated
saveRDS( climate_awareness_air, file: file.path("data-raw", "climate-panel_recoded.rds"))

What is Retrospective Survey Harmonization?

Thu, 04 Mar 2021 00:00:00 +0000

Reproducible ex post harmonization of survey microdata

Retrospective survey harmonization allows the comparison of opinion poll data conducted in different countries or time. In this example we are working with data from surveys that were ex ante harmonized to a certain degree – in our tutorials we are choosing questions that were asked in the same way in many natural languages. For example, you can compare what percentage of the European people in various countries, provinces and regions thought climate change was a serious world problem back in 2013, 2015, 2017 and 2019.

We developed the retroharmonize R package to help this process. We have tested the package with about 80 Eurobarometer, 5 Afrobarometer survey files extensively, and a bit with Arabbarometer files. This allows the comparison of various survey answers in about 70 countries. This policy-oriented survey programs were designed to be harmonized to a certain degree, but their ex post harmonization is still necessary, challenging and errorprone. Retrospective harmonization includes harmonization of the different coding used for questions and answer options, post-stratification weights, and using different file formats.

Eurobarometer, Afrobaromer, Arab Barometer and Latinobarómetro make survey files that are harmonized across countries available for research with various terms. Our retroharmonize is not affiliated with them, and to run our examples, you must visit their websites, carefully read their terms, agree to them, and download their data yourself. What we add as a value is that we help to connect their files across time (from different years) or across these programs.

The survey programs mentioned above publish their data in the proprietary SPSS format. This file format can be imported and translated to R objects with the haven package; however, we needed to re-design haven’s labelled_spss class to maintain far more metadata, which, in turn, a modification of the labelled class. The haven package was designed and tested with data stored in individual SPSS files.

The author of labelled, Joseph Larmarange describes two main approaches to work with labelled data, such as SPSS’s method to store categorical data in the Introduction to labelled.

Two main approaches of labelled data conversion.

Our approach is a further extension of Approach B. Survey harmonization in our case always means the joining data from several SPSS files, which requires a consistent coding among several data sources. This means that data cleaning and recoding must take place before conversion to factors, character or numeric vectors. This is particularly important with factor data (and their simple character conversions) and numeric data that occasionally contains labels, for example, to describe the reason why certain data is missing. Our tutorial vignette labelled_spss_survey gives you more information about this.

In the next series of tutorials, we will deal with an array of problems. These are not for the faint heart – you need to have a solid intermediate level of R to follow.

Tidy, joined survey data

The original files identifiers may not be unique, we have to create new, truly unique identifiers. Weighting may not be straightforward.
Neither the number of observations or the number of variables (which represents the survey questions and their translation to coded data) is the same. Certain data may be only present in one survey and not the other. This means that you will likely to run loops on lists and not data.frames, but eventually you must carefully join them.

Class conversion

Similar questions may be imported from a non-native R format, in our case, from an SPSS files, in an inconsistent manner. SPSS’s variable formats cannot be translated unambiguously to R classes. retroharmonize introduced a new S3 class system that handles this problem, but eventually you will have to choose if you want to see a numeric or character coding of each categorical variable.
The harmonized surveys, with harmonized variable names and harmonized value labels, must be brought to consistent R representations (most statistical functions will only work on numeric, factor or character data) and carefully joined into a single data table for analysis.

Harmonization of variables and variable labels

Same variables may come with dissimilar variable names and variable labels. It may be a challenge to match age with age. We need to harmonize the names of variables.
The harmonized variables may have different labeling. One may call refused answers as declined and the other refusal. On a simple choice, climate change may be ‘Climate change’ or Problem: Climate change. Binary choices may have survey-specific coding conventions. Value labels must be harmonized. There are good tools to do this in a single file - but we have to work with several of them.

Missing value harmonization

There are likely to be various types of missing values. Working with missing values is probably where most human judgment is needed. Why are some answers missing: was the question not asked in some questionnaires? Is there a coding error? Did the respondent refuse the question, or sad that she did not have an answer? retroharmonize has a special labeled vector type that retains this information from the raw data, if it is present, but you must make the judgment yourself – in R, eventually you will either create a missing category, or use NA_character_ or NA_real_.

That’s a lot to put on your plate.

It is unlikely that you will be able to work with completely unfamiliar survey programs if you do not have a strong intermediate level of R. Our package comes with tutorials for Eurobarometer, Afrobarometer and our development version already covers Arab Barometer, highlighting some peculiar issues with these survey programs, that we hope to give a head start for less experienced R users.