Data fusion
In this first notebook, we load the data taken from Google, explore and fuse them.
Loading all the data sets
> census = read.csv("census.csv.bz2", header = TRUE) > pollution = read.csv("pollution.weekly.csv.bz2", header = TRUE) > searches = read.csv("search.trends.csv.bz2", header = TRUE) > weather = read.csv("weather.week.csv.bz2", header = TRUE) > coords = read.csv("county.coordinates.csv.bz2", header = TRUE)
The raw data files are available from here:
- census.csv.bz2
- pollution.weekly.csv.bz2
- search.trends.csv.bz2
- weather.week.csv.bz2
- county.coordinates.csv.bz2
The census data
The census data contain 3064 observations and 20
variables, which are imported as follows by default.
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| fips_county_code | integer | 0 |
| sub_region_1 | character | 0 |
| sub_region_2 | character | 0 |
| bachelor_degree_or_higher | numeric | 0 |
| college_or_associate_degree | numeric | 0 |
| POVALL_2019 | integer | 0 |
| Unemployment_rate_2020 | numeric | 0 |
| Median_Household_Income_2019 | numeric | 0 |
| Metro_2013 | numeric | 0 |
| RUCC_2013 | integer | 0 |
| Pop | integer | 0 |
| vote_rate | numeric | 0 |
| Household_Size | numeric | 0 |
| Density | numeric | 0 |
| bachelor_degree_or_higher_b | integer | 0 |
| college_or_associate_degree_b | integer | 0 |
| POVALL_2019_p | numeric | 0 |
| POVALL_2019_b | integer | 0 |
| Unemployment_rate_2020_b | integer | 0 |
| Median_Household_Income_2019_b | integer | 0 |
The unique county identifier is fips_county_code. Each fips_county_code appears multiple
times in the data with different vote_rate records. Summing up the vote rates for each
fips_county_code and keeping only the unique records gives one entry for each county and vote rates that
are all strictly less than 1.
> # identify duplicate ids which appear multiple times. > ttt = table(census$fips_county_code) > dup_fips_id = names(ttt[ttt > 1]) > > for (id in dup_fips_id) { + # sum up the vote rates for each id. + all_votes = sum(census[census$fips_county_code == id, "vote_rate"]) + census[census$fips_county_code == id, "vote_rate"] = all_votes + }#FOR > > # keep only the unique records. > census = unique(census) > # double-check that each id appears only once. > ttt = table(census$fips_county_code) > length(ttt[ttt > 1])
[1] 0
> # double check that all vote rates are in [0, 1]. > summary(census$vote_rate)
Min. 1st Qu. Median Mean 3rd Qu. Max. 0.0873 0.5121 0.6316 0.6090 0.7258 0.9035
Vote rates should be normalised to a [0, 1] scale to match other continuous variables.
> census[, "vote_rate"] = census[, "vote_rate"] / 100
The variable sub_region_1 is the US state, and sub_region_2 is the county name. We cannot
use sub_region_2 for anything other than data exploration because there are counties with the same name in
different states.
> names(census)[names(census) == "sub_region_2"] = "county" > names(census)[names(census) == "sub_region_1"] = "state" > # an example: > census[census$county == "Adams County", c("county", "state")]
county state
279 Adams County Colorado
746 Adams County Illinois
802 Adams County Indiana
1334 Adams County Mississippi
1445 Adams County Nebraska
1916 Adams County Ohio
2149 Adams County Pennsylvania
3002 Adams County Wisconsin
Both these variables can be encoded as factors for convenience.
> census[, "county"] = as.factor(census[, "county"]) > census[, "state"] = as.factor(census[, "state"])
The bachelor_degree_or_higher_b and college_or_associate_degree_b identify the proportions
of the population with a specific educational achievement above or below a certain threshold.
The variables bachelor_degree_or_higher and college_or_associate_degree encode the
proportions of the population with the respective educational achievements.
> standard.histogram(census[, "bachelor_degree_or_higher"], + legend = "Proportion of the population with a bachelor degree or better")
> standard.histogram(census[, "college_or_associate_degree"], + legend = "Proportion of the population with a college/associate degree")
Both percentages can be normalised to a [0, 1] scale to match other variables.
> census[, "bachelor_degree_or_higher"] = census[, "bachelor_degree_or_higher"] / 100 > census[, "college_or_associate_degree"] = census[, "college_or_associate_degree"] / 100
The bachelor_degree_or_higher_b and college_or_associate_degree_b are indicators that
identify the proportions of the population with a specific educational achievement that are above or below a certain
threshold.
> range(census[census[, c("bachelor_degree_or_higher_b")] == 0, + c("bachelor_degree_or_higher")])
[1] 0.05386811 0.29220734
> range(census[census[, c("bachelor_degree_or_higher_b")] == 1, + c("bachelor_degree_or_higher")])
[1] 0.2923395 0.7529932
> range(census[census[, c("college_or_associate_degree_b")] == 0, + c("college_or_associate_degree")])
[1] 0.1124183 0.3325122
> range(census[census[, c("college_or_associate_degree_b")] == 1, + c("college_or_associate_degree")])
[1] 0.3325383 0.4733221
> standard.xyplot(bachelor_degree_or_higher ~ seq(nrow(census)), + data = census, xlab = "", ylab = "Proportion with a bachelor degree or higher", + col = c("skyblue", "orange")[census[, c("bachelor_degree_or_higher_b")] + 1])
> standard.xyplot(college_or_associate_degree ~ seq(nrow(census)), + data = census, xlab = "", ylab = "Proportion with a college/associate degree", + col = c("skyblue", "orange")[census[, c("college_or_associate_degree_b")] + 1])
They are best encoded as factors.
> census[, c("bachelor_degree_or_higher_b")] = + factor(census[, c("bachelor_degree_or_higher_b")], levels = c(0, 1), + labels = c("LOW", "HIGH")) > census[, c("college_or_associate_degree_b")] = + factor(census[, c("college_or_associate_degree_b")], levels = c(0, 1), + labels = c("LOW", "HIGH"))
The variable POVALL_2019 is the number of people of all ages in poverty. It comes both as an absolute
value and a proportion under the name POVALL_2019_p. The variable POVALL_2019_b then
identifies the proportions (not the absolute values!) above a certain threshold.
> range(census[census[, "POVALL_2019_b"] == 0, "POVALL_2019_p"])
[1] 0.03370641 0.17268516
> range(census[census[, "POVALL_2019_b"] == 1, "POVALL_2019_p"])
[1] 0.1726897 0.3401949
> standard.xyplot(POVALL_2019_p ~ seq(nrow(census)), + data = census, xlab = "", ylab = "POVALL", + col = c("skyblue", "orange")[census[, c("POVALL_2019_b")] + 1])
There is no point in keeping POVALL_2019 in the data when we have both the corresponding proportion and
the population size. The proportion should also be rescaled to [0, 1] for consistency with other continuous
variables.
> all.equal(census[, "POVALL_2019"] / census[, "Pop"], census[, "POVALL_2019_p"])
[1] TRUE
> census[, names(census) == "POVALL_2019"] = NULL > census[, "POVALL_2019_p"] = census[, "POVALL_2019_p"] / 100 > census[, c("POVALL_2019_b")] = + factor(census[, c("POVALL_2019_b")], levels = c(0, 1), labels = c("LOW", "HIGH")) > names(census)[names(census) == "POVALL_2019_p"] = "POVALL" > names(census)[names(census) == "POVALL_2019_b"] = "POVALL_b"
The variable Unemployment_rate_2020 is the percentage of unemployed people in the county, and
Unemployment_rate_2020_b is the corresponding binarised transform that identifies unemployment rates above
a certain threshold.
> standard.histogram(census[, "Unemployment_rate_2020"], + legend = "Unemployment Rate (2020)")
> standard.xyplot(Unemployment_rate_2020 ~ seq(nrow(census)), + data = census, xlab = "", ylab = "Unemployment rate", + col = c("skyblue", "orange")[census[, c("Unemployment_rate_2020_b")] + 1])
We should rescale the rate to [0, 1] for consistency, rename it for brevity, and transform the binarised variable to a factor.
> census[, "Unemployment_rate_2020"] = census[, "Unemployment_rate_2020"] / 100 > census[, c("Unemployment_rate_2020_b")] = + factor(census[, c("Unemployment_rate_2020_b")], levels = c(0, 1), + labels = c("LOW", "HIGH")) > names(census)[names(census) == "Unemployment_rate_2020"] = "unemployment" > names(census)[names(census) == "Unemployment_rate_2020_b"] = "unemployment_b"
The variable Median_Household_Income_2019, in dollars, and its binarised transform
Median_Household_Income_2019_b.
> standard.histogram(log10(census[, "Median_Household_Income_2019"]), + legend = "Median Household Income (2019, on a log10 scale)")
> standard.xyplot(Median_Household_Income_2019 ~ seq(nrow(census)), + data = census, xlab = "", ylab = "Median Household Income", + col = c("skyblue", "orange")[census[, c("Median_Household_Income_2019_b")] + 1])
Household income does not appear to correlate with the unemployment rate, but it does correlate with the proportion of university graduates.
> standard.xyplot(Median_Household_Income_2019 ~ unemployment, + data = census, xlab = "Unemployment", ylab = "Median Household Income")
> standard.xyplot(Median_Household_Income_2019 ~ bachelor_degree_or_higher, + data = census, xlab = "bachelor_degree_or_higher", ylab = "Median Household Income", + regression = TRUE)
Shortening the variable names for brevity. We should also change the income to a log10() scale for
consistency with other numeric variables and transform the binarised version into a factor.
> census[, "Median_Household_Income_2019"] = + log10(census[, "Median_Household_Income_2019"]) > census[, "Median_Household_Income_2019_b"] = + factor(census[, c("Median_Household_Income_2019_b")], levels = c(0, 1), + labels = c("LOW", "HIGH")) > names(census)[names(census) == "Median_Household_Income_2019"] = "household_income" > names(census)[names(census) == "Median_Household_Income_2019_b"] = "household_income_b"
Metro_2013 distinguishes between metropolitan and non-metropolitan counties. The variable
RUCC_2013 contains the Rural-Urban Continuum Codes (RUCC), which further classifies how rural/urban a
county is. RUCC is a score from 1 to 9, defined as follows:
- Metropolitan counties
- 1: Counties in metro areas of 1 million population or more.
- 2: Counties in metro areas of 250,000 to 1 million population.
- 3: Counties in metro areas of fewer than 250,000 population.
- Non-metropolitan counties
- 4: Urban population of 20,000 or more, adjacent to a metro area.
- 5: Urban population of 20,000 or more, not adjacent to a metro area.
- 6: Urban population of 2,500 to 19,999, adjacent to a metro area.
- 7: Urban population of 2,500 to 19,999, not adjacent to a metro area.
- 8: Completely rural or less than 2,500 urban population, adjacent to a metro area.
- 9: Completely rural or less than 2,500 urban population, not adjacent to a metro area.
> table(census[, c("Metro_2013", "RUCC_2013")])
RUCC_2013
Metro_2013 1 2 3 4 5 6 7 8 9
0 0 0 0 213 91 345 173 16 13
1 346 327 268 0 0 0 0 0 0
Both variables are best encoded as factors.
> census[, "Metro_2013"] = + factor(census[, "Metro_2013"], levels = c(0, 1), labels = c("metro", "non-metro")) > census[, "RUCC_2013"] = factor(census[, "RUCC_2013"], levels = as.character(1:9)) > names(census)[names(census) == "Metro_2013"] = "Metro" > names(census)[names(census) == "RUCC_2013"] = "RUCC"
Pop is the county population size.
> standard.histogram(log10(census[, "Pop"]), legend = "Population (on a log10 scale)")
We should put it on a log10() scale for consistency with other numeric variables.
> census[, "Pop"] = log10(census[, "Pop"]) > names(census)[names(census) == "Pop"] = "population"
Household_Size is the average number of people living together in each household.
> standard.histogram(census[, "Household_Size"], legend = "Household size")
Renamed for consistency with household_income.
> names(census)[names(census) == "Household_Size"] = "household_size"
Density is the population density, most likely in people/sqft.
> standard.histogram(log10(census[, "Density"]), legend = "Population Density (on a log10 scale)")
Renamed for clarity and reascaled to log10() for consistency with other numeric variables.
> census[, "Density"] = log10(census[, "Density"]) > names(census)[names(census) == "Density"] = "population_density"
The final form of the
census data set.
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| fips_county_code | integer | 0 |
| state | factor | 0 |
| county | factor | 0 |
| bachelor_degree_or_higher | numeric | 0 |
| college_or_associate_degree | numeric | 0 |
| unemployment | numeric | 0 |
| household_income | numeric | 0 |
| Metro | factor | 0 |
| RUCC | factor | 0 |
| population | numeric | 0 |
| vote_rate | numeric | 0 |
| household_size | numeric | 0 |
| population_density | numeric | 0 |
| bachelor_degree_or_higher_b | factor | 0 |
| college_or_associate_degree_b | factor | 0 |
| POVALL | numeric | 0 |
| POVALL_b | factor | 0 |
| unemployment_b | factor | 0 |
| household_income_b | factor | 0 |
The pollution data
The pollution data contain 1742937 observations and
22 variables, which are imported as follows by default:
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| state_code | integer | 0.000 |
| county_code | integer | 0.000 |
| fips_county_code | integer | 0.000 |
| week_start_date | character | 0.000 |
| weekly_avg_CO | numeric | 0.799 |
| weekly_max_CO | numeric | 0.799 |
| weekly_aqi_CO | numeric | 0.799 |
| weekly_avg_NO2 | numeric | 0.781 |
| weekly_max_NO2 | numeric | 0.781 |
| weekly_aqi_NO2 | numeric | 0.781 |
| weekly_avg_SO2 | numeric | 0.683 |
| weekly_max_SO2 | numeric | 0.683 |
| weekly_aqi_SO2 | numeric | 0.683 |
| weekly_avg_O3 | numeric | 0.445 |
| weekly_max_O3 | numeric | 0.445 |
| weekly_aqi_O3 | numeric | 0.445 |
| weekly_avg_PM25 | numeric | 0.581 |
| weekly_max_PM25 | numeric | 0.581 |
| weekly_aqi_PM25 | numeric | 0.581 |
| weekly_avg_PM10 | numeric | 0.549 |
| weekly_max_PM10 | numeric | 0.549 |
| weekly_aqi_PM10 | numeric | 0.549 |
Variables are already homogeneous: they comprise a triplet of mean, maximum and air quality index (AQI) for each pollutant. The ranges are very different for different pollutants.
> bwplot(variable ~ value, data = melt(pollution[, -(1:4)]), + scales = list(x = list(log = 10)))
Therefore, we only need to remove the county_code and state_code, which are redundant given
the census data, and to shorten the variable names for brevity.
> pollution[, names(pollution) %in% c("county_code", "state_code")] = NULL > names(pollution) = sub("^weekly_", "", names(pollution))
For convenience and to match other data sets, we rename the date of the first day of the week.
> names(pollution)[names(pollution) == "week_start_date"] = "week"
Finally, we rename the pollutants to follow the same naming pattern as the weather variables.
> names(pollution) = gsub("avg_(.*)", "\\1_mean", names(pollution)) > names(pollution) = gsub("max_(.*)", "\\1_max", names(pollution)) > names(pollution) = gsub("aqi_(.*)", "\\1_aqi", names(pollution))
The air quality index for a pollutant is defined as the mean concentration divided by the national standard and multiplied by 100. Therefore, it correlates almost perfectly with the mean concentrations.
> cor(pollution$CO_mean, pollution$CO_aqi, use = "pairwise.complete")
[1] 0.9718042
> cor(pollution$NO2_mean, pollution$NO2_aqi, use = "pairwise.complete")
[1] 0.9518696
> cor(pollution$SO2_mean, pollution$SO2_aqi, use = "pairwise.complete")
[1] 0.8392479
> cor(pollution$O3_mean, pollution$O3_aqi, use = "pairwise.complete")
[1] 0.8365865
> cor(pollution$PM10_mean, pollution$PM10_aqi, use = "pairwise.complete")
[1] 0.9856848
> cor(pollution$PM25_mean, pollution$PM25_aqi, use = "pairwise.complete")
[1] 0.9606402
> pollution[, grepl("_aqi", names(pollution))] = NULL
The final form of the
pollution data set.
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| fips_county_code | integer | 0.000 |
| week | character | 0.000 |
| CO_mean | numeric | 0.799 |
| CO_max | numeric | 0.799 |
| NO2_mean | numeric | 0.781 |
| NO2_max | numeric | 0.781 |
| SO2_mean | numeric | 0.683 |
| SO2_max | numeric | 0.683 |
| O3_mean | numeric | 0.445 |
| O3_max | numeric | 0.445 |
| PM25_mean | numeric | 0.581 |
| PM25_max | numeric | 0.581 |
| PM10_mean | numeric | 0.549 |
| PM10_max | numeric | 0.549 |
Google searches for the medical conditions
The searches data contain 327758 observations and 13
variables, which are imported as follows by default:
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| country_region_code | character | 0 |
| country_region | character | 0 |
| sub_region_1 | character | 0 |
| sub_region_1_code | character | 0 |
| sub_region_2 | character | 0 |
| fips_county_code | integer | 0 |
| place_id | character | 0 |
| date | character | 0 |
| symptom_anxiety | numeric | 0 |
| symptom_depression | numeric | 0 |
| symptom_atopic_dermatitis | numeric | 0 |
| symptom_obesity | numeric | 0 |
| symptom_sleep_disorder | numeric | 0 |
First, we rename the date.
> names(searches)[names(searches) == "date"] = "week"
Then we remove the location variables we do not need to merge the data sets.
> searches[, c("country_region_code", "country_region", "sub_region_1")] = NULL > searches[, c("sub_region_1_code", "sub_region_2", "place_id")] = NULL
Finally, we rename the conditions for brevity.
> names(searches) = gsub("^symptom_", "", names(searches))
The weather data
The weather data contain 327920 observations and 18
variables, which by default are imported as:
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| statefp | integer | 0.000 |
| countyfp | integer | 0.000 |
| countyns | integer | 0.000 |
| county_name | character | 0.000 |
| fips_county_code | integer | 0.000 |
| date | character | 0.000 |
| temp_mean | numeric | 0.009 |
| temp_sum | numeric | 0.000 |
| temp_max | numeric | 0.009 |
| temp_min | numeric | 0.009 |
| wdsp_mean | numeric | 0.009 |
| wdsp_sum | numeric | 0.000 |
| wdsp_max | numeric | 0.009 |
| wdsp_min | numeric | 0.009 |
| prcp_mean | numeric | 0.009 |
| prcp_sum | numeric | 0.000 |
| prcp_max | numeric | 0.009 |
| prcp_min | numeric | 0.009 |
We can remove all redundant location variables, leaving only fips_county_code.
> weather$statefp = weather$countyfp = weather$county_name = weather$countyns = NULL
We also rename the date to be consistent with the other data sets and give more descriptive names to the other variables.
> names(weather)[names(weather) == "date"] = "week" > names(weather) = gsub("^temp_", "temperature_", names(weather)) > names(weather) = gsub("^wdsp_", "windspeed_", names(weather)) > names(weather) = gsub("^prcp_", "precipitations_", names(weather))
For obvious reasons, the _mean and _sum variables are perfectly correlated up to
measurement noise: we only keep the former. Before we remove the sums over the whole week, we use them to fill in the
missing values in the means.
> cor(weather$windspeed_sum, weather$windspeed_mean, use = "pairwise.complete")
[1] 0.9954285
> cor(weather$precipitations_sum, weather$precipitations_mean, use = "pairwise.complete")
[1] 0.9924422
> cor(weather$temperature_sum, weather$temperature_mean, use = "pairwise.complete")
[1] 0.96267
> weather[is.na(weather$windspeed_mean), "windspeed_mean"] = + weather[is.na(weather$windspeed_mean), "windspeed_sum"] / 7 > weather[is.na(weather$precipitations_mean), "precipitations_mean"] = + weather[is.na(weather$precipitations_mean), "precipitations_sum"] / 7 > weather[is.na(weather$temperature_mean), "temperature_mean"] = + weather[is.na(weather$temperature_mean), "temperature_sum"] / 7 > > weather[, grepl("_sum", names(weather))] = NULL
A quick overview of the weather variables.
> bwplot(variable ~ value, data = melt(weather[, -(1:2)]))
The final form of the
weather data set.
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| fips_county_code | integer | 0.000 |
| week | character | 0.000 |
| temperature_mean | numeric | 0.000 |
| temperature_max | numeric | 0.009 |
| temperature_min | numeric | 0.009 |
| windspeed_mean | numeric | 0.000 |
| windspeed_max | numeric | 0.009 |
| windspeed_min | numeric | 0.009 |
| precipitations_mean | numeric | 0.000 |
| precipitations_max | numeric | 0.009 |
| precipitations_min | numeric | 0.009 |
The county coordinates
The coords data contain 3233 observations and 4
variables, which are imported as follows by default:
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| fips_code | integer | 0 |
| name | character | 0 |
| lng | numeric | 0 |
| lat | numeric | 0 |
The only variable that is redundant is the county name; the other variables could use names that are consistent with other data sets or just more descriptive.
> coords = coords[, names(coords) != "name"] > names(coords) = c("fips_county_code", "longitude", "latitude")
Merging all the data sets
The variables that act as keys to merge the different data sets are fips_county_code and
week. Not all codes appear in all data sets, and neither do all weeks.
> all.weeks = sort(unique(c(pollution$week, searches$week, weather$week))) > pollution$week = ordered(pollution$week, levels = all.weeks) > searches$week = ordered(searches$week, levels = all.weeks) > weather$week = ordered(weather$week, levels = all.weeks)
- The
pollutiondata ranges from1990-01-01to2023-05-01. - The
searchesdata ranges from2020-03-02to2023-10-23. - The
weatherdata ranges from2020-01-06to2023-11-06. - The
censusdata has no time dimension. - The county coordinates obviously have no time dimension.
> all.fips = unique(c(pollution$fips_county_code, + searches$fips_county_code, + census$fips_county_code, + weather$fips_county_code, + coords$fips_county_code)) > pollution$fips_county_code = factor(pollution$fips_county_code, levels = all.fips) > searches$fips_county_code = factor(searches$fips_county_code, levels = all.fips) > census$fips_county_code = factor(census$fips_county_code, levels = all.fips) > weather$fips_county_code = factor(weather$fips_county_code, levels = all.fips) > coords$fips_county_code = factor(coords$fips_county_code, levels = all.fips)
- The
pollutiondata covers1470counties. - The
searchesdata covers2021counties. - The
weatherdata covers1652counties. - The
censusdata covers1792counties. - The
coordsdata covers3233counties.
> ps = merge(pollution, searches, by = c("fips_county_code", "week"), all = TRUE) > psc = merge(census, ps, by = "fips_county_code", all = TRUE) > pscw = merge(psc, weather, by = c("fips_county_code", "week"), all = TRUE) > all.data = merge(pscw, coords, by = "fips_county_code", all = TRUE)
The final merged data set contains 2053067 observations and 48
variables.
We restrict ourselves to the period and the counties where we observe the conditions we would like to study.
> # we should observe at least half of the conditions and half of the pollutants. > missing.pollutants = + rowSums(is.na(all.data[, c("NO2_mean", "SO2_mean", "O3_mean", "PM25_mean", "PM10_mean")])) > missing.conditions = + rowSums(is.na(all.data[, c("anxiety", "depression", "atopic_dermatitis", "obesity", "sleep_disorder")])) > > useful.data = all.data[missing.pollutants <= 3 & missing.conditions <= 3, ] > > # drop CO as it is missing > 50% of the time. > useful.data = useful.data[, setdiff(colnames(useful.data), c("CO_mean", "CO_max"))]
This reduces the number of observations to 53797 over
566 counties and
136 weeks.
Finally, we can fill in the missing county and state labels. Both variables come from the
census data, and we have counties for which we have no census data, but we do have pollutants.
> patch = data.frame( + code = c("2090", "4013", "6001", "6013", "6027", "6037", "6043", "6051", + "6059", "6065", "6067", "6071", "6073", "6077", "6085", "12011", + "12057", "12086", "12095", "12099", "17031", "25017", "26125", + "26163", "27053", "32003", "36005", "36061", "36081", "36103", + "39035", "39049", "42003", "42101", "46127", "48029", "48113", + "48201", "48439", "48453", "49013", "49035", "51059", "53033", + "56023", "36085", "37173", "51071", "2170", "55003", "2110"), + name = c("Fairbanks North Star", "Maricopa", "Alameda", "Contra Costa", "Inyo", + "Los Angeles", "Mariposa", "Mono", "Orange", "Riverside", "Sacramento", + "San Bernardino", "San Diego", "San Joaquin", "Santa Clara", + "Broward", "Hillsborough", "Miami-Dade", "Orange", "Palm Beach", + "Cook", "Middlesex", "Oakland", "Wayne", "Hennepin", "Clark", + "Bronx", "New York", "Queens", "Suffolk", "Cuyahoga", "Franklin", + "Allegheny", "Philadelphia", "Union", "Bexar", "Dallas", "Harris", + "Tarrant", "Travis", "Duchesne", "Salt Lake", "Fairfax", "King", + "Lincoln", "Richmond", "Swain", "Giles", "Matanuska-Susitna", + "Ashland", "Juneau"), + state = c("Alaska", "Arizona", rep("California", 13), rep("Florida", 5), + "Illinois", "Massachusetts", rep("Michigan", 2), "Minnesota", + "Nevada", rep("New York", 4), rep("Ohio", 2), rep("Pennsylvania", 2), + "South Dakota", rep("Texas", 5), rep("Utah", 2), "Virginia", + "Washington", "Wyoming", "New York", "North Carolina", "Virginia", + "Alaska", "Wisconsin", "Alaska") + ) > > for (i in seq(nrow(patch))) { + if (!(patch$name[i] %in% levels(useful.data$county))) + levels(useful.data$county) = c(levels(useful.data$county), patch$name[i]) + useful.data[useful.data$fips_county_code == patch$code[i], "county"] = patch$name[i] + useful.data[useful.data$fips_county_code == patch$code[i], "state"] = patch$state[i] + }#FOR
| LABEL | TYPE | MISSING VALUES |
|---|---|---|
| fips_county_code | factor | 0.000 |
| week | ordered | 0.000 |
| state | factor | 0.000 |
| county | factor | 0.000 |
| bachelor_degree_or_higher | numeric | 0.086 |
| college_or_associate_degree | numeric | 0.086 |
| unemployment | numeric | 0.086 |
| household_income | numeric | 0.086 |
| Metro | factor | 0.086 |
| RUCC | factor | 0.086 |
| population | numeric | 0.086 |
| vote_rate | numeric | 0.086 |
| household_size | numeric | 0.086 |
| population_density | numeric | 0.086 |
| bachelor_degree_or_higher_b | factor | 0.086 |
| college_or_associate_degree_b | factor | 0.086 |
| POVALL | numeric | 0.086 |
| POVALL_b | factor | 0.086 |
| unemployment_b | factor | 0.086 |
| household_income_b | factor | 0.086 |
| NO2_mean | numeric | 0.555 |
| NO2_max | numeric | 0.555 |
| SO2_mean | numeric | 0.532 |
| SO2_max | numeric | 0.532 |
| O3_mean | numeric | 0.135 |
| O3_max | numeric | 0.135 |
| PM25_mean | numeric | 0.112 |
| PM25_max | numeric | 0.112 |
| PM10_mean | numeric | 0.502 |
| PM10_max | numeric | 0.502 |
| anxiety | numeric | 0.000 |
| depression | numeric | 0.000 |
| atopic_dermatitis | numeric | 0.000 |
| obesity | numeric | 0.000 |
| sleep_disorder | numeric | 0.000 |
| temperature_mean | numeric | 0.143 |
| temperature_max | numeric | 0.145 |
| temperature_min | numeric | 0.145 |
| windspeed_mean | numeric | 0.143 |
| windspeed_max | numeric | 0.145 |
| windspeed_min | numeric | 0.145 |
| precipitations_mean | numeric | 0.143 |
| precipitations_max | numeric | 0.145 |
| precipitations_min | numeric | 0.145 |
| longitude | numeric | 0.000 |
| latitude | numeric | 0.000 |
The observations are spread across different weeks as follows.
The observations are spread across different counties as follows. Observations from outlying territories
(Puerto Rico, Northern Mariana Islands, American Samoa) are excluded from the plot and appear as NA in the
legend. The largest possible number is the number of weeks still present in the data
(136).
Mon Apr 28 13:54:54 2025 with bnlearn
5.0-20240418
and R version 4.3.2 (2023-10-31).
