Research project idea (500 words). Here you will describe the idea for your research project, including some background on the topic, a description of your data source(s) and a description of your variable(s). You will also state your research questions and goals related to your project.
In December 2014, Minneapolis, Minnesota passed a city ordinance allowing for the construction of accessory dwelling units (ADUs, also known as “granny flats”) on existing single-family lots. [1] The ordinance was the first in a series of steps to address the costs of housing in the city by increasing housing supply. In 2018, Minneapolis continued this approach by banning single-family zoning, thus requiring a 3-unit minimum on each lot for new construction.
The 2014 ordinance allows single-family homeowners to construct ADUs on their property. Theoretically, this should add housing supply to the city and apply downward pressure on housing costs. Critics argued ADUs wouldn’t make a significant impact on housing supply since they are relatively expensive to build (around $100,000 each at the time) and the property owner had to continue to reside in either the existing home or the ADU. In other words, they would primarily be used by family members and would not enter the general rental market.
The 2018 ordinance banning single-family zoning was one of the first of its kind. Single-family zoning is highly valued by homeowners but advocates of high-density housing argue it should be phased out. It is highly unusual for existing single-family zoning to be modified, so Minneapolis is entering uncharted territory for housing policy.
This research project will analyze changes in rental affordability in Minneapolis between 2014, when the ADU ordinance was passed, and 2020. Comparing housing costs before and after the ADU ordinance and single-family zoning ban will provide evidence of the effects of these changes in housing policy. The research question is how did rental affordability change in Minneapolis, Minnesota, five years after the city passed an ordinance allowing ADUs on single family lots in December 2014 and two years after the city banned single family zoning in 2018?
I will map 3 variables from the U.S. Census Bureau American Community Survey Public Use Microdata (2014 and 2020) related to housing costs in Minneapolis, Minnesota, in 2014 and 2020, by census tract. The variables are percent of renters in each tract (derived from the TEN variable, housing tenure), the percent of renters who are cost burdened (derived from GRPIP, gross rental as a percentate of household income), and the change in percent of renters who are cost burdened from 2014 to 2020. “Cost burdened” is defined as paying greater than 30% of income towards housing costs.
I will produce the following maps and tables:
References
Description of data and GIS processes (250-500 words, 1-2 tables, 2-3 figures). Here you will describe your data, including the source and origin, as well as a plan for what GIS operations you will be conducting during the course of your project.
The source of data is the U.S. Census Bureau’s American Community Survey Public Use Microdata from 2014 and 2019. I will access the data via the get_pums() function in r. The variables I will use are the following:
For the GIS operations, I will download the shapefiles using the purrr::map() function and merge the data with the Minneapolis housing data. I will produce the following maps:
First, let’s download and clean Minnesota PUMS data for 2014.
mnpums2014 <- get_pums(
variables = c("PUMA", "TYPE", "TEN", "GRPIP", "HHT", "RELP"),
state = "MN",
variables_filter = list(SPORDER = 1, TYPE = 1), # SPORDER = 1 gets households, TYPE = 1 gets housing units and eliminates group quarters.
#puma = c(1405, 1406, 1407), # I can't figure out how to select by puma. Maybe capitalize "PUMA"?
survey = "acs1",
year = 2014
)Getting data from the 2014 1-year ACS Public Use Microdata SampleThis is a sample of all respondents in Minnesota for 2014. Let’s check the sample size.
nrow(mnpums2014)[1] 21524Let’s create a new dataframe with only the PUMAs that cover Minneapolis (1405, 1406, and 1407) in 2014.
dat2014 <- mnpums2014 %>%
filter(PUMA %in% c("1405", "1406", "1407"))Let’s modify the PUMA variable in the 2014 dataframe so the PUMA is 5 digits.
dat2014 <- dat2014 %>%
mutate(PUMA = case_when(.$PUMA == "1405" ~ "01405",
.$PUMA == "1406" ~ "01406",
.$PUMA == "1407" ~ "01407",
)
)Let’s check the sample size.
nrow(dat2014)[1] 1023Let’s check the distribution by PUMA.
tabyl(dat2014$PUMA) dat2014$PUMA n percent
01405 344 0.3362659
01406 331 0.3235582
01407 348 0.3401760In order to get just one observation per household, we need to ensure RELP == 0 for every record.
tabyl(dat2014$RELP) dat2014$RELP n percent
0 1023 1So now we know there is only one observation per household. Let’s see the distribution of renters vs. non-renters. 1 is owned with a mortgage, 2 is owned free and clear, 3 is rented, 4 is occupied without payment of rent.
tabyl(dat2014$TEN) dat2014$TEN n percent
1 401 0.391984360
2 158 0.154447703
3 461 0.450635386
4 3 0.002932551Let’s recode the TEN variable categories to reflect renters and non-renters.
dat2014 <- dat2014 %>%
mutate(tenure = (ifelse(.$TEN == 3, 'Renter', 'Non-renter')))
tabyl(dat2014$tenure) dat2014$tenure n percent
Non-renter 562 0.5493646
Renter 461 0.4506354Making a data frame with percent of renters in 2014
renters2014 <- dat2014%>%
group_by(PUMA, tenure) %>%
summarise(N = sum(WGTP)) %>%
mutate(Proportion = N/sum(N)) %>%
filter(tenure == 'Renter')`summarise()` has grouped output by 'PUMA'. You can override using the
`.groups` argument.renters2014$Proportion <- renters2014$Proportion*100
renters2014# A tibble: 3 × 4
# Groups: PUMA [3]
PUMA tenure N Proportion
<chr> <chr> <dbl> <dbl>
1 01405 Renter 25280 47.7
2 01406 Renter 21451 40.9
3 01407 Renter 40292 60.8Let’s filter only renters and check the distribution of rent as a percentage of income.
dat2014 <- dat2014 %>%
filter(tenure == "Renter")
dat2014$GRPIP <- (as.numeric(dat2014$GRPIP))/100
hist(dat2014$GRPIP)
Let’s create a new variable called “cost_burden” with three levels:
dat2014 <- dat2014 %>%
mutate(cost_burden = case_when(.$GRPIP <.30 ~ "Not cost burdened",
.$GRPIP >=.30 & .$GRPIP <.50 ~ "Cost burdened",
.$GRPIP >=.50 ~ "Extremely cost burdened",
)
)Let’s check the distribution of rent cost burden in Minneapolis in 2014.
tabyl(dat2014$cost_burden) dat2014$cost_burden n percent
Cost burdened 95 0.2060738
Extremely cost burdened 136 0.2950108
Not cost burdened 230 0.4989154Now, let’s download and clean Minnesota PUMS data for 2019. (PUMS data was not released for 2020 and PUMS geographies are not available for 2021, so those years are not available.)
mnpums2019 <- get_pums(
variables = c("PUMA", "TYPE", "TEN", "GRPIP", "HHT", "RELSHIPP"),
state = "MN",
variables_filter = list(SPORDER = 1, TYPE = 1), # SPORDER = 1 gets households, TYPE = 1 gets housing units and eliminates group quarters.
#puma = c(1405, 1406, 1407), # I can't figure out how to select by puma. Maybe capitalize "PUMA".
survey = "acs1",
year = 2019
)Getting data from the 2019 1-year ACS Public Use Microdata SampleThis is a sample of PUMS respondents in Minnesota for 2019. Let’s check the sample size.
nrow(mnpums2019)[1] 22576First, let’s create a new dataframe with only the PUMAs that cover Minneapolis (1405, 1406, and 1407) in 2019.
dat2019 <- mnpums2019 %>%
filter(PUMA %in% c("01405", "01406", "01407"))Let’s check the sample size.
nrow(dat2019)[1] 1054Let’s check the distribution by PUMA.
tabyl(dat2019$PUMA) dat2019$PUMA n percent
01405 350 0.3320683
01406 336 0.3187856
01407 368 0.3491461In order to get just one observation per household, we need to ensure RELSHIPP == 20 for every record.
tabyl(dat2019$RELSHIPP) dat2019$RELSHIPP n percent
20 1054 1So now we know there is only one observation per household. Let’s see the distribution of renters vs. non-renters. 1 is owned with a mortgage, 2 is owned free and clear, 3 is rented, 4 is occupied without payment of rent.
tabyl(dat2019$TEN) dat2019$TEN n percent
1 413 0.3918406
2 174 0.1650854
3 461 0.4373814
4 6 0.0056926Let’s recode the TEN variable categories to reflect renters and non-renters.
dat2019 <- dat2019 %>%
mutate(tenure = (ifelse(.$TEN == 3, 'Renter', 'Non-renter')))
tabyl(dat2019$tenure) dat2019$tenure n percent
Non-renter 593 0.5626186
Renter 461 0.4373814Making a data frame with percent of renters in 2019.
renters2019 <- dat2019%>%
group_by(PUMA, tenure) %>%
summarise(N = sum(WGTP)) %>%
mutate(Proportion = N/sum(N)) %>%
filter(tenure == 'Renter')`summarise()` has grouped output by 'PUMA'. You can override using the
`.groups` argument.renters2019$Proportion <- renters2019$Proportion*100
renters2019# A tibble: 3 × 4
# Groups: PUMA [3]
PUMA tenure N Proportion
<chr> <chr> <dbl> <dbl>
1 01405 Renter 27542 49.5
2 01406 Renter 23169 42.8
3 01407 Renter 44253 59.6Let’s filter only renters and check the distribution of rent as a percentage of income.
dat2019 <- dat2019 %>%
filter(tenure == "Renter")
dat2019$GRPIP <- (as.numeric(dat2019$GRPIP))/100
hist(dat2019$GRPIP)
Let’s create a new variable called “cost_burden” with three levels:
dat2019 <- dat2019 %>%
mutate(cost_burden = case_when(.$GRPIP <.30 ~ "Not cost burdened",
.$GRPIP >=.30 & .$GRPIP <.50 ~ "Cost burdened",
.$GRPIP >=.50 ~ "Extremely cost burdened",
)
)Let’s check the distribution of rent cost burden in Minneapolis in 2019.
tabyl(dat2019$cost_burden) dat2019$cost_burden n percent
Cost burdened 86 0.1865510
Extremely cost burdened 94 0.2039046
Not cost burdened 281 0.6095445Now I need to merge the 2014 and 2019 dataframes.
Let’s rename RELSHIPP as RELP in the 2019 dataframe.
dat2019 <- dat2019 %>%
rename_at('RELSHIPP', ~'RELP')Let’s add a year column to both dataframes.
dat2014 <- dat2014 %>%
mutate(year = "2014")
dat2019 <- dat2019 %>%
mutate(year = "2019")Let’s merge these dataframes into one.
dat <- rbind(dat2014, dat2019)
head(dat)# A tibble: 6 × 14
SERIALNO WGTP PWGTP PUMA TEN GRPIP HHT RELP SPORDER TYPE ST tenure
<chr> <dbl> <dbl> <chr> <chr> <dbl> <chr> <chr> <chr> <chr> <chr> <chr>
1 2339 96 96 01405 3 0.17 6 0 1 1 27 Renter
2 10042 106 106 01407 3 0.24 7 0 1 1 27 Renter
3 22566 132 132 01405 3 0.31 3 0 1 1 27 Renter
4 25036 350 350 01405 3 0.22 3 0 1 1 27 Renter
5 27547 319 319 01406 3 0.21 3 0 1 1 27 Renter
6 30186 178 178 01407 3 0.97 6 0 1 1 27 Renter
# ℹ 2 more variables: cost_burden <chr>, year <chr>Let’s make a table showing the change in rent cost burden in the city from 2014 to 2019. This code weighs the data with WGTP, the variable for household weight.
tab1 <- dat %>%
group_by(year, cost_burden) %>%
summarise(N = sum(WGTP)) %>%
mutate(Proportion = N/sum(N))`summarise()` has grouped output by 'year'. You can override using the
`.groups` argument.gt(tab1)| cost_burden | N | Proportion |
|---|---|---|
| 2014 | ||
| Cost burdened | 17160 | 0.1971892 |
| Extremely cost burdened | 28138 | 0.3233398 |
| Not cost burdened | 41725 | 0.4794709 |
| 2019 | ||
| Cost burdened | 16108 | 0.1696222 |
| Extremely cost burdened | 20488 | 0.2157449 |
| Not cost burdened | 58368 | 0.6146329 |
This table shows the share of cost burdened renters declined from 19.7% in 2014 to 17.0% in 2019. The proportion of “Extremely cost burdened” renters declined from 32.2% in 2014 to 21.6% in 2019. The share of “Not cost burdened” renters rose from 47.9% in 2014 to 61.5% in 2019.
Let’s create a new variable called “cost_burden2” with two levels:
dat <- dat %>%
mutate(cost_burden2 = case_when(.$GRPIP <.30 ~ "Not cost burdened",
.$GRPIP >=.30 ~ "Cost burdened",
)
)Let’s make a table showing the change in rent cost burden (cost_burden2) in the city from 2014 to 2019. This code weighs the data with WGTP, the variable for household weight.
tab2 <- dat %>%
group_by(PUMA, year, cost_burden2) %>%
summarise(N = sum(WGTP)) %>%
mutate(Proportion = N/sum(N))`summarise()` has grouped output by 'PUMA', 'year'. You can override using the
`.groups` argument.gt(tab2)| cost_burden2 | N | Proportion |
|---|---|---|
| 01405 - 2014 | ||
| Cost burdened | 14819 | 0.5861946 |
| Not cost burdened | 10461 | 0.4138054 |
| 01405 - 2019 | ||
| Cost burdened | 12676 | 0.4602425 |
| Not cost burdened | 14866 | 0.5397575 |
| 01406 - 2014 | ||
| Cost burdened | 12346 | 0.5755443 |
| Not cost burdened | 9105 | 0.4244557 |
| 01406 - 2019 | ||
| Cost burdened | 9089 | 0.3922914 |
| Not cost burdened | 14080 | 0.6077086 |
| 01407 - 2014 | ||
| Cost burdened | 18133 | 0.4500397 |
| Not cost burdened | 22159 | 0.5499603 |
| 01407 - 2019 | ||
| Cost burdened | 14831 | 0.3351411 |
| Not cost burdened | 29422 | 0.6648589 |
Downloading the shapefile for Minnesota for 2014. I tried to use the FIPS code for Hennepin County, but the shape file is for all of Minnesota.
# What package is the map function from (purrr)? Why are we calling tigris::pumas? What is 'cb'? Can I use this sf file with tmap?
hmap2014 <- map("27053", # Getting PUMA geography for Hennepin County.
tigris::pumas,
class = "sf",
cb = TRUE,
year = 2014) %>%
reduce(rbind) # This changes the list to a dataframe. left_join won't work if you don't do this.
hcpumas2014 <- as.vector(hmap2014$PUMACE10)Downloading the shapefile for Minnesota for 2019. I tried to use the FIPS code for Hennepin County, but the shape file is for all of Minnesota.
hmap2019 <- map("27053",
tigris::pumas,
class = "sf",
cb = TRUE,
year = 2019) %>%
reduce(rbind)
hcpumas2019 <- as.vector(hmap2019$PUMACE10)
# Cartographic boundary PUMAs are not yet available for years after 2019. Use the argument `year = 2019` instead to request your data.Use mapview to view the shapefile. It shows all the PUMAs in the state of Minnesota.
mapview(hmap2014)Filter for the 3 PUMAs I am interested in for 2014 and create a quick map.
mlps2014_sf <- hmap2014 %>%
filter(PUMACE10 %in% c('01405', '01406', '01407'))
mapview(mlps2014_sf)Should I expand the mapping/analysis to all of Hennepin County?
Using ggplot to make a map showing percent of renters in 2014.
# These PUMAS are missing the 0 in the front.
map1 <- mlps2014_sf %>%
left_join(renters2014, by = c("PUMACE10" = "PUMA")) %>%
ggplot(aes(fill = Proportion)) +
geom_sf() +
scale_fill_viridis_b(
name = NULL,
option = "magma",
labels = scales::label_percent(1)
) +
# labs(title = "Mean Gross Rent as a Percentage of Houshold Income \nMapped by Public Use Microdata Area") +
ggtitle("Percent of households who rent, 2014",
subtitle = "Source: U.S. Census Bureau, 2014 ACS 1 year") +
theme_void()
map1
Using ggplot to make a map showing percent of renters in 2019.
map2 <- mlps2014_sf %>%
left_join(renters2019, by = c("PUMACE10" = "PUMA")) %>%
ggplot(aes(fill = Proportion)) +
geom_sf() +
scale_fill_viridis_b(
name = NULL,
option = "magma",
labels = scales::label_percent(1)
) +
# labs(title = "Mean Gross Rent as a Percentage of Houshold Income \nMapped by Public Use Microdata Area") +
ggtitle("Percent of households who rent, 2019",
subtitle = "Source: U.S. Census Bureau, 2014 ACS 1 year") +
theme_void()
map2
Next, use ggplot to make maps of the percent who are cost burdened in 2014 and 2019
Make a dataframe grouped by PUMA showing percent of renters who are cost burdened in 2014.
cb2014 <- dat %>%
filter(year == 2014) %>%
group_by(PUMA, cost_burden2) %>%
summarise(N = sum(WGTP)) %>%
mutate(Proportion = N/sum(N)) %>%
filter(cost_burden2 == 'Cost burdened')`summarise()` has grouped output by 'PUMA'. You can override using the
`.groups` argument.cb2014$Proportion <- cb2014$Proportion*100
cb2014# A tibble: 3 × 4
# Groups: PUMA [3]
PUMA cost_burden2 N Proportion
<chr> <chr> <dbl> <dbl>
1 01405 Cost burdened 14819 58.6
2 01406 Cost burdened 12346 57.6
3 01407 Cost burdened 18133 45.0Join my 2014 cost burden data with the shapefile and make a map.
Make a dataframe grouped by PUMA showing percent of renters who are cost burdened in 2019.
cb2019 <- dat %>%
filter(year == 2019) %>%
group_by(PUMA, cost_burden2) %>%
summarise(N = sum(WGTP)) %>%
mutate(Proportion = N/sum(N)) %>%
filter(cost_burden2 == 'Cost burdened')`summarise()` has grouped output by 'PUMA'. You can override using the
`.groups` argument.cb2019$Proportion <- cb2019$Proportion*100
cb2019# A tibble: 3 × 4
# Groups: PUMA [3]
PUMA cost_burden2 N Proportion
<chr> <chr> <dbl> <dbl>
1 01405 Cost burdened 12676 46.0
2 01406 Cost burdened 9089 39.2
3 01407 Cost burdened 14831 33.5Join my 2019 cost burden data with the shapefile and make a map.
Preliminary results (500-750 words, 1-2 tables, 1-2 figures). This is where you will describe the preliminary results of your project. Here you will describe the justification of which techniques you used, and describe the preliminary results of your analysis based on the procedures you conducted.
gt(renters2014)| tenure | N | Proportion |
|---|---|---|
| 01405 | ||
| Renter | 25280 | 47.66394 |
| 01406 | ||
| Renter | 21451 | 40.89254 |
| 01407 | ||
| Renter | 40292 | 60.82179 |
In 2014, 48% of households in PUMA 1405 were renters. 40.1% of households in PUMA 1406 were renters. 60.1% of households in PUMA 1407 were renters.
tmap1 <- mlps2014_sf %>%
left_join(renters2014, by = c("PUMACE10" = "PUMA"))
tm_shape(tmap1)+
tm_polygons("Proportion",
title="Percent of renters in each PUMA",
palette="Reds"
)+
tm_format("World",
title="Percent of renters in 2014",
legend.format = list(digits = 2),
legend.outside=T)+
tm_scale_bar()+
tm_compass()
gt(renters2019)| tenure | N | Proportion |
|---|---|---|
| 01405 | ||
| Renter | 27542 | 49.52706 |
| 01406 | ||
| Renter | 23169 | 42.80805 |
| 01407 | ||
| Renter | 44253 | 59.64981 |
In 2019, 49.6% of households in PUMA 1405 were renters. 42.8% of households in PUMA 1406 were renters. 59.6% of households in PUMA 1407 were renters.
tmap2 <- mlps2014_sf %>%
left_join(renters2019, by = c("PUMACE10" = "PUMA"))
tm_shape(tmap2)+
tm_polygons("Proportion",
title="Percent of renters in each PUMA",
palette="Reds"
)+
tm_format("World",
title="Percent of renters in 2019",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
gt(cb2014)| cost_burden2 | N | Proportion |
|---|---|---|
| 01405 | ||
| Cost burdened | 14819 | 58.61946 |
| 01406 | ||
| Cost burdened | 12346 | 57.55443 |
| 01407 | ||
| Cost burdened | 18133 | 45.00397 |
In 2014, 58.6% of renters were cost burdened in PUMA 1405. 57.6% of renters were cost burdened in PUMA 1406. 45.0% of renters were cost burdened in PUMA 1407.
tmap3 <- mlps2014_sf %>%
left_join(cb2014, by = c("PUMACE10" = "PUMA"))
tm_shape(tmap3)+
tm_polygons("Proportion",
title="Percent of renters who are cost burdened",
palette="Reds"
)+
tm_format("World",
title="Percent who are cost burdened in 2014",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
gt(cb2019)| cost_burden2 | N | Proportion |
|---|---|---|
| 01405 | ||
| Cost burdened | 12676 | 46.02425 |
| 01406 | ||
| Cost burdened | 9089 | 39.22914 |
| 01407 | ||
| Cost burdened | 14831 | 33.51411 |
In 2019, 46.0% of renters were cost burdened in PUMA 1405. 39.2% of renters were cost burdened in PUMA 1406. 33.6% of renters were cost burdened in PUMA 1407.
tmap4 <- mlps2014_sf %>%
left_join(cb2019, by = c("PUMACE10" = "PUMA"))
tm_shape(tmap4)+
tm_polygons("Proportion",
title="Percent of renters who are cost burdened",
palette="Reds"
)+
tm_format("World",
title="Percent who are cost burdened in 2019",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
Create a data frame to find the difference in cost burden between 2014 and 2019.
fiveyr <- cb2019 %>%
select(PUMA, Proportion)
fiveyr <- rename(fiveyr, Prop2019 = Proportion)
fiveyr <- merge(fiveyr, cb2014)
fiveyr <- fiveyr %>%
select(PUMA, Prop2019, Proportion)
fiveyr <- rename(fiveyr, Prop2014 = Proportion)
fiveyr <- fiveyr %>%
mutate(diff = round(((Prop2019 - Prop2014)), digits = 1))
gt(fiveyr)| PUMA | Prop2019 | Prop2014 | diff |
|---|---|---|---|
| 01405 | 46.02425 | 58.61946 | -12.6 |
| 01406 | 39.22914 | 57.55443 | -18.3 |
| 01407 | 33.51411 | 45.00397 | -11.5 |
Between 2014 and 2019, the percent of renters who were costs burdened declined by 12.6% in PUMA 1405.
The proportion of renters who were cost burdened declined by 18.3% in PUMA 1406.
The proportion of renters who were cost burdened declined by 11.5% in PUMA 1407.
tmap5 <- mlps2014_sf %>%
left_join(fiveyr, by = c("PUMACE10" = "PUMA"))
tm_shape(tmap5)+
tm_polygons("diff",
title="Change in renters who are cost burdened from 2014 to 2019",
palette="Reds"
)+
tm_format("World",
title="Change in renters who are cost burdened from 2014 to 2019",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
In this project, I analyzed the change in affordability in Minneapolis, Minnesota at the PUMA level. I downloaded microdata from the US Census Bureau and analyzed rent as a percentage of income. “Cost burdened” was defined as paying greater than 30% of income on housing. I calculated summary statistics for the three PUMAS in Minneapolis. The city is comprised of three PUMAs, 1405, 1406, and 1407.
The preliminary results are:
Final project discussion (500-750 words, tables and figures as necessary). Here you will describe the overall results of your project, including how your results aligned with your research questions from post 1, the overall takeaway from your project, the limitations of your project and how you could build upon this project in the future.
On December 5, 2014, the Minneapolis City Council passed a zoning code amendment which allows accessory dwelling units (or ADUs) to be constructed on lots with single or two-family homes (“Accessory Dwelling Unit Zoning Code Text Amendment Passes in Minneapolis,” 2014). An ADU must provide a separate living space with a separate entrance. They are also known as granny flats, mother-in-law apartments, carriage houses, or casitas (Speck, 2018).
ADUs emerged as a policy option to increase housing supply and provide affordable options in areas impacted by housing shortages (Kim et al., 2023). ADUs can be created by converting portions of existing homes, constructing additions to existing homes, or constructing new structures on the same lots as existing structures (Speck, 2018). One benefit of ADUs is that they can add housing units to built-up areas, adding both new units to the supply of housing as well as providing diversity in the types of housing available.
However, ADUs are controversial in some areas due to homeowner resistance, regulatory burden, or the cost of construction (Speck, 2018). For example, many ADU ordinances require that the unit be built with off-street parking. This increases the expense for construction and takes up additional space on existing lots, which can disincentivize ADU construction (Brinig & Garnett, 2013). Due to their controversial nature, only a few municipalities nationwide have passed ADU ordinances (Kim et al., 2023; Maaoui, 2018).
One of the main goals of ADU ordinances is to improve housing affordability for renters (Speck, 2018). However, few studies have analyzed the effect of ADU ordinances on affordability. Kim et al. (2023) utilized multilevel logistic regression to examine the effects of an ADU ordinance in Los Angeles, California. They found the types of properties that constructed ADUs after the ordinance was passed were more varied in terms of lot characteristics. For example, the ordinance probably led to more ADU development on lots near bus transit. However, the study did not explore the relationship between the ordinance and changes in rent costs.
Gabbe (2019) analyzed zoning changes to address high housing prices in Los Angeles, between 2016 and 2020. The zoning changes included easing regulations on ADUs. The study concluded Los Angeles made progress in five categories of policy recommendations, including ADUs, but the study explicitly did not evaluate the effect on rent costs.
Maaoui (2018) analyzed the correlation between minority household concentration and the permitting of ADUs in unincorporated areas of King County, Washington. The study found permits were positively correlated with Black and Latino households relative to White households, and negatively correlated with Asian households. Furthermore, permits were negatively correlated with low- and high-income households, and positively correlated with middle income households. Again, the study did not investigate the effect of ADUs on rent costs.
This study contributes to the literature by calculating whether the percentage of renters who are cost burdened changed in Minneapolis between 2014, when the ADU ordinance was passed, and 2019. The year 2019 was chosen because it was the last year 1-year PUMS data was available prior to the COVID epidemic. Due to the COVID epidemic, the U.S. Census Bureau did not release standard 1-year PUMS data for 2020 (U.S. Census Bureau, 2022a).
The data source for this study is the American Community Survey Public Use Microdata Sample Files, also known as ACS microdata or PUMS (U.S. Census Bureau, 2021). ACS microdata consists of records with information about characteristics of individual households, with personalized information removed (U.S. Census Bureau, 2021). One-year PUMS include records for about 1 percent of the total population. The samples are divided by Public Use Microdata Areas (PUMAs). Each PUMA has a population of at least 100,000 and no more than 200,000.
The benefit of microdata is that it provides individual responses to the full range of topics in the ACS. PUMS contains approximately 200 housing-level variables, which permits data users to analyze specific population groups and create custom variables that are not available through ACS summary tables. This allows more flexibility than ACS summary data, allowing users to focus on demographic characteristics and create unique household variables.
The geographic area of analysis for this project is the city of Minneapolis, Minnesota. In order to extract a sample for the city of Minneapolis from PUMS, the PUMAs covering Minneapolis were identified as Minnesota PUMAs 1405, 1406, and 1407 (U.S. Census Bureau, 2022b). Samples of 1-year PUMS data were extracted for the years of 2014 and 2019 using the get_pums() function in R.
The objective of this study is to measure the change in cost burden in Minneapolis, Minnesota, between the years of 2014 and 2019. “Rent cost” is defined as rent as a percentage of household income. “Cost burden” is defined as the percent of renters whose rent exceeds 30% of their gross income for the past 12 months.
The PUMS variable GRPIP is defined as “Gross rent as a percentage of household income past 12 months” (U.S. Census Bureau, 2015). It is a 3-digit numerical variable with 001 to 100 representing 1% to 100% and the number 101 representing 101% or more. The proportion of cost burdened renters was derived from GRPIP for the three PUMAs.
Additional variables were extracted in order to conduct the analysis. The PUMS variable TEN is defined as “Tenue.” It is a one-digit numerical variable with five possible values. The value “3” reflects that the household is “rented.” This variable was used to isolate renters from homeowners.
“Rent cost” is defined as rent as a percentage of income. In 2014, the median rent cost for renters in Minneapolis was 30.0%. In 2019, the median was 25.0%. Between 2014 and 2019, the median decreased by 5.0%, meaning rent as a proportion of income decreased during that time.
The following five maps were produced to show changes between 2014 and 2019: - Map 1: Percent of all households who are renters in each census tract in 2014 - Map 2: Percent of all households who are renters in each census tract in 2019 - Map 3: Percent of renters in each census tract who are cost burdened in 2014 - Map 4: Percent of renters in each census tract who are cost burdened in 2019 - Map 5: Change in percent of renters who are cost burdened from 2014 to 2019
A series of statistical tests were performed to determine if the mean difference in rent costs between the two years is statistically significant. First, the assumption of normality was tested for the two samples. A Shapiro-Wilk test for normality was conducted using the shapiro.test() function in R Studio. The p-value was less than 0.05 for both values, so the assumption of normality was rejected. Next, Boxplots were produced to visually show outliers. The data showed outliers in the 2019 data.
The lack of normality and presence of outliers violate the assumptions of a t-test. As an alternate, a Wilcoxson Rank Sum test was performed using the wilcox.test() function in R Studio. This tests for the differences in the sample means when violations of normality and outliers are present.
gt(fiveyr)| PUMA | Prop2019 | Prop2014 | diff |
|---|---|---|---|
| 01405 | 46.02425 | 58.61946 | -12.6 |
| 01406 | 39.22914 | 57.55443 | -18.3 |
| 01407 | 33.51411 | 45.00397 | -11.5 |
The percent of renters who are cost burden declined in every PUMA between 2014 and 2019. In PUMA 1405, the percent of cost burdened renters declined by 12.6%. In PUMA 1406, the percent of cost burdened renters declined by 18.3%. In PUMA 1407, the percent of cost burdened renters declined by 11.5%.
The Wilcoxson Rank Sum test of the median rent cost by year resulted in a two-sided test p-value = 7.892e-05. This indicates that we should reject the null hypothesis that distributions are equal and conclude that there is a statistically significant difference in rent costs between the two years. Unweighted descriptive statistics using the summarise() function indicated that the median rent cost in 2014 was 30% and the median rental cost in 2019 was 26%. The difference between the median rent costs of each year is about 4%. We are 95% certain that the median difference between 2014 and 2019 is between 2% and 7%. Thus, the 4% decline in unweighted media rent cost from 2014 to 2019 is statistically significant.
The study has a number of limitations. Minneapolis has a relatively small population and is covered by only three PUMAs. As a result, the sample sizes for some demographic groups are relatively small which could have affected the calculations. Studying a population with a larger sample size, such as Hennepin County, may produce more accurate descriptive statistics. The challenge with this approach is that changes in policy often occur at smaller geographic levels, such as the city level. Therefore, small sample sizes create a challenge for researchers studying the impacts of policies on smaller geographies.
A test for statistical significance was only conducted for the sample of all renters, but were not conducted for individual PUMAs. Furthermore, the study was not designed to control for other variables or find evidence of causation. Other statistical methods, such as multivariate regression or difference-in-difference could be used to analyze rent costs in relationship with other demographic variables.
The scope of the research could be increased to larger geographical areas or multiple geographical areas. Increasing the geographic scope would increase the sample sizes and improve the accuracy of the statistical tests. Future research could use regression models or tests such as difference-in-difference to explore the impact of the ADU ordinance in the context of multiple variables. Regression models could include demographic characteristics such as sex, race/ethnicity, age groups, income quantiles, presence of children, and citizenship status. Other variables could include local, state, or national housing policies, as well as indexes for inequality. Similar research could be conducted on other policies intended to decrease rent costs, such as incentives to construct multifamily housing and “upzoning” neighborhoods with single family homes. Furthermore, physical features related to rent affordability could be mapped and anaylzed in GIS, including sites of ADU construction.
Projecting the maps using EPSG 2812, a projection for Minnesota South. Correting the scales so the colors and legends are consistent between maps.
new_tmap1 <- st_transform(tmap1, crs = 2812)
new_tmap2 <- st_transform(tmap2, crs = 2812)
new_tmap3 <- st_transform(tmap3, crs = 2812)
new_tmap4 <- st_transform(tmap4, crs = 2812)
new_tmap5 <- st_transform(tmap5, crs = 2812)tm_shape(new_tmap1)+
tm_polygons(title="Percent of renters in each PUMA",
palette="Reds",
col = "Proportion",
breaks = c(40, 45, 50, 55, 60, 65)
)+
tm_format("World",
title="Percent of renters in 2014",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
tm_shape(new_tmap2)+
tm_polygons(title="Percent of renters in each PUMA",
palette="Reds",
col = "Proportion",
breaks = c(40, 45, 50, 55, 60, 65)
)+
tm_format("World",
title="Percent of renters in 2019",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
tm_shape(new_tmap3)+
tm_polygons(title="Percent of renters who are cost burdened",
palette="Blues",
col = "Proportion",
breaks = c(30, 35, 40, 45, 50, 55, 60)
)+
tm_format("World",
title="Percent who are cost burdened in 2014",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
tm_shape(new_tmap4)+
tm_polygons(title="Percent of renters who are cost burdened",
palette="Blues",
col = "Proportion",
breaks = c(30, 35, 40, 45, 50, 55, 60)
)+
tm_format("World",
title="Percent who are cost burdened in 2019",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
tm_shape(new_tmap5)+
tm_polygons(title="Change in renters who are cost burdened from 2014 to 2019",
palette="Greens",
col = "diff",
breaks = c(-20, -18, -16, -14, -12, -10)
)+
tm_format("World",
title="Change in renters who are cost burdened from 2014 to 2019",
legend.outside=T)+
tm_scale_bar()+
tm_compass()
Performing the Shapiro-Wilk Test for Normality on the variable GRPIP in the samples from 2014 and 2019.
swtab <- dat %>%
group_by(year) %>%
summarise(`W Statistic` = shapiro.test(GRPIP)$statistic,
`p-value` = shapiro.test(GRPIP)$p.value)
swtab# A tibble: 2 × 3
year `W Statistic` `p-value`
<chr> <dbl> <dbl>
1 2014 0.844 6.08e-21
2 2019 0.825 4.56e-22Because the p-value < 0.05 for both samples, we reject the assumption of normality. The distributions are not normal.
ggplot(dat, aes(x = year, y = GRPIP, fill = year)) +
stat_boxplot(geom ="errorbar", width = 0.5) +
geom_boxplot(fill = "light blue") +
stat_summary(fun=mean, geom="point", shape=10, size=3.5, color="black") +
ggtitle("Boxplots of rent cost burdent in 2014 and 2019") +
theme_bw() + theme(legend.position="none")
The boxplots show many outliers in 2019.
The lack of normality or severe impact of outliers can violate independent sample t-test assumptions and ultimately the results. If this happens, there are other options.
Performing a nonparametric Mann-Whitney U test is the most popular alternative. This is also known as the Mann-Whitney-Wilcoxon or the Wilcoxon Rank Sum test. This test is considered robust to violations of normality and outliers (among others) and tests for differences in mean ranks.
dat %>%
select(GRPIP, year) %>%
group_by(year) %>%
summarise(n = n(),
median=median(GRPIP, na.rm = TRUE),
mean = mean(GRPIP, na.rm = TRUE),
sd = sd(GRPIP, na.rm = TRUE),
stderr = sd/sqrt(n),
LCL = mean - qt(1 - (0.05 / 2), n - 1) * stderr,
UCL = mean + qt(1 - (0.05 / 2), n - 1) * stderr,
min=min(GRPIP, na.rm = TRUE),
max=max(GRPIP, na.rm = TRUE),
IQR=IQR(GRPIP, na.rm = TRUE))# A tibble: 2 × 11
year n median mean sd stderr LCL UCL min max IQR
<chr> <int> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 2014 461 0.3 0.410 0.297 0.0138 0.383 0.437 0 1.01 0.35
2 2019 461 0.26 0.341 0.261 0.0122 0.317 0.365 0 1.01 0.25The following assumptions must be met in order to run a Mann-Whitney U test:
The dependent response variable is median rent cost as a percentage of income.
The independent categorical variable is year.
test1<-wilcox.test(GRPIP ~ year, data=dat, na.rm=TRUE, paired=FALSE, exact=FALSE, conf.int=TRUE)
print(test1)
Wilcoxon rank sum test with continuity correction
data: GRPIP by year
W = 122212, p-value = 7.892e-05
alternative hypothesis: true location shift is not equal to 0
95 percent confidence interval:
0.02007798 0.06992372
sample estimates:
difference in location
0.04005175 The rent cost in each year is not normally distributed. Outliers exist in each group. A Mann-Whitney U test is more appropriate than a traditional independent samples t-test to compare the rent cost between the two years.
The Mann-Whitney U test results in a two-sided test p-value = 7.892e-05. This indicates that we should reject the null hypothesis that distributions are equal and conclude that there is a significant difference in rent cost between the two years. Descriptive statistics indicate that the median cost burden in 2014 was 30% and the median rent cost in 2019 was 26%. The difference between the median rent cost of each year is about 4%. We are 95% certain that the median difference between 2014 and 2019 is between 2% and 7%. Thus, the 4% decline in rent cost from 2014 to 2019 is statistically significant.
Accessory Dwelling Unit zoning Code Text Amendment Passes in Minneapolis. (2014, December 5). Targeted News Service. ProQuest One Academic. https://login.libweb.lib.utsa.edu/login?url=https://www.proquest.com/wire-feeds/accessory-dwelling-unit-zoning-code-text/docview/1632227679/se-2?accountid=7122
Brinig, M. F., & Garnett, N. S. (2013). A Room of One’s Own? Accessory Dwelling Unit Reforms and Local Parochialism. The Urban Lawyer, 45(3), 519–569. JSTOR.
Gabbe, C. J. (2019). Changing Residential Land Use Regulations to Address High Housing Prices: Evidence From Los Angeles. Journal of the American Planning Association, 85(2), 152–168. https://doi.org/10.1080/01944363.2018.1559078
Kim, D., Baek, S.-R., Garcia, B., Vo, T., & Wen, F. (2023). The influence of accessory dwelling unit (ADU) policy on the contributing factors to ADU development: An assessment of the city of Los Angeles. Journal of Housing and the Built Environment. https://doi.org/10.1007/s10901-022-10000-2
Maaoui, M. (2018). A granny flat of one’s own? The households that build accessory-dwelling units in Seattle’s King County. Berkeley Planning Journal, 30(1).
Speck, J. (2018). Encourage Granny Flats: Allow and Incentivize Accessory Dwelling Units. In J. Speck, Walkable City Rules (pp. 28–29). Island Press/Center for Resource Economics. https://doi.org/10.5822/978-1-61091-899-2_12
U.S. Census Bureau. (2015). 2014 ACS PUMS Data Dictionary.
U.S. Census Bureau. (2021). Understanding and Using the American Community Survey Public Use Microdata Sample Files: What Data Users Need to Know. US Government Printing Office. https://www.census.gov/content/dam/Census/library/publications/2021/acs/acs_pums_handbook_2021.pdf
U.S. Census Bureau. (2022a, March 28). 2020 PUMS Data. https://www.census.gov/programs-surveys/acs/microdata/access/2020.html
U.S. Census Bureau. (2022b, August 22). Public Use Microdata Areas (PUMAs). https://www.census.gov/programs-surveys/geography/guidance/geo-areas/pumas.html