Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment [Attachment
I Source of Somerville's research and summary of results] 2 number of distribution of establishment with employment range] 3 Source of Somerville's research and summary of results] 4 Data source, year, geography land of aggregation summary] 5 CBP data record layout] 6 WRLURI data record layout] 7 Descriptive summary of individual data of WRLURI] 8 Sub index components] 9 Example of Geographical code matching process] data set, 140 MSA records with variables name] 11 Descriptive statistics of counties and MSAs while data processing] 12 Index for census data used in correlation test] 13 Correlation test with size of firm]
List of Tables and Pictures
. 16 [Table 1 Summary of assumptions explaining firm size].. [Table 2 Somerville's modeling in organization structure].. 25 [Table 3 County Business Pattern data set coverage and industry code summary]. 28 [Table 4 Summary of employment and establishment].. 29 [Table 5 Summary statistic of number of permit and construction cost]. [Table 6 Fair market rent of 2003, 2005, and 2007 data set.. [Table 7 Developable land area and fragmentation of land, 2002 and 2007].. [Table 8 Summary of jurisdictions in Wharton regulatory data set]. [Table 9 Summary statistic of components of factor loading regulatory measurement]. 39 [Table 10 W RLURI summary statistics]... [Table 11 Summary Statistic of sub index]... [Table 12 Summary of place records linking to county level].. 43 [Table 13 County Base average number of employees per establishment summary].. 47 [Table counties' average number of employees per establishment]. 47. 49 [Table 15 Sub index table with its demographic features].. [Table 16 Correlation test of firm size with demographic treats of MSAs]. 53 [Table 17 Correlation between firm size and rent].. 55 [Table 18 Correlation test result]... 55 [Table 19 Regression Summary table to refine previous model]. 57 [Table 20 Correlation test in model 0]... 58. 59 [Table 21 Correlation test in Model 1]... [Table 22 sensitivity test in model 0 (2005) and Somerville's 1987 data results].. 59 [Table 23 Correlation test after adding median house value].. 61 [Table 24 Regression results with 5 years housing price index of 2003, 2005, and 2007]. 62 [Table 25 Regression results with yearly housing price index over 5 years]. 63 [Table 26 Results summary of each regulatory variables impact to model 5].. 65 [Table 27 Regression Summary table with regulation data]... 65 [Table 28 Regression results after adding regional dummies].. 66 [Table 29 Sensitivity test by grouping the MSA's size and price].. 68 [Table 30 Comparison of regression Results by existence of density restriction].. 71 [Table 31 Comparison of regression Results by intensity of Approval process]. 72 [Table 32 Results summary based on Model 5]... 73 [Table 33 Rent level and regulatory sub index relevance, based on 2005 county level]. 76 [Picture 1 Summary of research question with needed test in Theory].. [Picture 2 Somervile's framework to integrate key features in housing supply]. [Picture 3 Structure of dependent variables to explain firm size spatial variation]. [Picture 4 Structure of variables and measurement to assess firm size spatial variation].. [Picture 5 Estimation of fair market rent based on two-bedroom survey. [Picture 6 Distribution of firm size based on county boundary, 2005]. [Picture 7. WRLURI distribution based on place level].. 48

[Graph 2 New starts of single-family housing from 1986 to
The data from 1998 to 2003 is missing, because the operative builders' survey data is integrated into residential construction builders, which group all contractors into one industrial code. The industrial code was changed from SIC to NAICS in 1998.
graph 2. Compared to the volatile movement of newly constructed housing stock, the constant number of establishments made us infer that there could be more volatility in the unit supply.
establishments Average of Errployment per of counties (operative builders)
The builders seem to adjust to the market by rapidly changing
their size, not by establishing new firms or closing existing
ones (Graph 36). Compared to the constant level of number of builders' establishments, the
[Graph 3 Average number of employees per establishment by county over 20 years.]
size shows more volatility than the actual fluctuation of new housing stock. One standard deviation per mean value of employees per establishment is 0.301, greater than one with starts of 0.17.
According to Graph 3 and Graph 2, the fluctuation in firm size was followed by a pattern of housing starts. Firm size decreased from 2004, and starts decreased from 2005. The number of establishments started to increase in 2006. There are some lags between housing supply and establishments and size of firms. The notable increase of establishments from 2006 to 2007 was followed by the increase of new housing starts and in the number of employees per establishment. Therefore, we can infer that the builders' size as the number of employees per establishments changes first to satisfy market demand and that changes of number of firms comes later. The size of firms seems more responsive to market condition
More detailed distribution of firm's size of each year is shown in attachments 1, using box plots.
than changes of number establishments by entering or existing market are. The size of firms, as measured by employees, shows that builders' behavior is related to the housing supply, when the number of establishments is highly inelastic. Therefore, size of firms seems good measure to understand builder's behavior. However, there are points of contention in this assumption, so more rigorous research on direct causality between size and market clearing process in demand is needed. However, because the dynamics of demand and supply equilibrium is not within the scope of this research, I will depend on the classical theory about the role of firm size.

5. Size Distribution of Business Firms The classic theory of firm size explains a unique size distribution based on minimization of cost factor over long-run operation. This theory proceeds under the assumption of product market competition. In contradiction to the theory, most changes in product demand are met by changes in firm size, not by entry or exit of firms. Moreover, firm growth appears to be independent of firm size. Finally, Simon, in collaboration with Bonini (1958) and Ijiri (1964), observed that by examining the distribution of firms by size at a single point in time, one can make inferences about the process governing firm growth and confirm that firm growth is independent of size. As mentioned earlier, I will assume that the builders have been actively synchronized with housing demand by adjusting their size. Historical pattern of homebuilder's size supports that understanding the determinants of builders' size as an indicator of the housing supply is important.
Chapter III: Research Design We will test each perspective's impact on firm size based on the comparison of markets in the U.S. metropolitan statistical area (MSA). The MSA is the most inclusive single market and uses a more rigorous statistical methodology.
1. Methodology I: Cross-sectional Comparison of Builder Size over Regions Comparison between MSA markets will be tested with a regression model for the years of 2002, 2005, and 2007. Cross-sectional comparison based on MSA level, depends on the homebuilder's monopolistic characteristics. Builder groups have traditionally been treated as the subject of supply under a perfect-competition paradigm. Researchers cite the large number of builders and the small share of the national market held by the county's largest builders to support this assumption. For example, surveys of the National Association of Homebuilders (NAHB) indicate that the 50 largest builders in the U.S. accounted for less than 10% of total national single-family starts. However, recent research has found a richer variation in industrial structure across markets. Somerville has argued that the industrial organization of homebuilders is more complex and varied has traditionally been argued. He showed that instead of a homogeneous competitive industry, there is a rich variation in market structure across metropolitan areas.

9 Summary of data source, year, and geography land of aggregation is shown in attachment 4.
10 Detailed layout of county business pattern dataset is shown in attachment 5. " New housing operative builders are primarily engaged in building new homes on land that the builder owns or controls. The land is included with the sale of the home. Establishments in this industry build single and/or multifamily homes. These establishments are referred to as merchant builders, production, or for-sale builders.
Therefore, there is issue for disclosure of data which captured as zero or data suppression flag. To handle this, I exclude zero value, which means there are no operative builders in that county or the data was not disclosed. Table 4 compares data description with and without zero value in the 2005 data set. A layout of yearly data is attached. Generally, establishment number does not show zero value. However, some establishments avoid revealing data about their employees. [Table 4 Summary of employment and establishment] Establishment Employment All Records Except 'O' All Records 104 Mean 8 Std. Dev. 0 Min 6029 Max 2213 Observation In 2005, the mean number of employees in each of the 2213 counties was 104.

dstribufion of firm rze

The distribution of firm size of each MSAs which calculated by aggregating counties within

7e class

MSAs depicted in Graph 5.
The firm size variation is quite
[Graph MSA base number of employees per large. The distribution is rightestablishment distribution.] skewed, meaning that there is wider size variation among large firms. Around 69% of establishments are within 0.5 standard deviations from mean firm size and 83% of firms are within one standard deviation from the mean firm size value. Seventeen percent of firms are more than one standard deviation from mean value.
2. Measuring Independent Variables 2.1 Measuring Demand For variables measuring demand, I will use single-family permits to describe aggregate demand, because they are available for a large cross section of MSAs. Though equilibrium quantity does not equal demand, the cross-MSA variation in housing permits will reflect differences across these MSAs in aggregate demand. Changes to the status of buildings often take place after the permit has been issued. Therefore, as actual equilibrium quantity between supply and demand, housing starts can be a more accurate measurement. However, there is no cross-MSA variation in housing starts. In addition, completions can be different from the number of starts. Based on the Census Bureau's survey on permits, starts, and completion, starts of single-family units were 2.5% greater than the number of permits. The number of completions of single-family units was 3.5% less than the number of starts. Therefore, the actual amount of supply under the assumption of equilibrium is 1.0% less than the number of permits. Therefore, permits seem not to be biased significantly as a measurement of market demand.

2.2 Measuring construction cost Data source is manufacturing, mining and construction statistics' MSA level statistics for yearly building permits. Because the available permit data is organized at the MSA level, it is important to manipulate the data with same MSA definitions and corresponding counties. I focus on the years after 2003 with the 2003 MSA definition by the OMB. One way to measure construction cost is to use the construction cost index. The index is
based on a sample of single housing units sold or under construction. However, construction cost indexes have performed poorly, due to the incorrect measurement of labor costs and a failure to address the endogeneity of construction costs and construction activity. Therefore, in this research, I will use aggregated construction value in counties. Construction cost per unit can be calculated by newly added single-family construction value divided by single-family permit in same data set in Census Bureau. I will focus on newly added units' reported construction value rather than construction prices in the region. Because the purpose of this research is to find the elements which affect the spatial variation of firm size, measurement for construction value which represents regional difference seems more proper than cost index. Summary statistics are presented below. [Table 5 Summary statistic of number of permit and construction cost] Number Average Number of unit Average total construction cost in each MSAs (in thousands dollars) permitted in each MSAs of Year Year of 3&4 MSAs lunit 2unit 5- unit lunit 2unit 3&4 unit 5- unit

535230

2.3 Measuring Housing Price For housing price, I use median housing value in census data, Fair Market Rent (FMR) 12 by Department of Housing and Urban Development, and housing price index by federal housing financial agency. Both fair market rent and housing price index are estimated by each institution rather than an actual measurement of the market. However, both are
Fair Market Rents (FMRs) are used to determine payment standard amounts for the Housing Choice Voucher program, The U.S. Department of Housing and Urban Development (HUD) annually estimates
FMRs for 530 metropolitan areas and 2,045 nonmetropolitan county FMR areas.

2.4 Measuring land supply and land assembly effect There were no explicit measures of the supply of land for development. Instead, we use variables that measure the total amount of land that can be readily developed and the extent to which land assembly is likely to be a problem. Farm and ranch land and MSA should be a rough measure of the supply of land that can be developed into lots for residential construction. Agricultural data on farm and ranch land in an MSA describes the supply of land suitable for development. While all MSAs have undeveloped land on the outskirts of the urban area, much of this land can be costly to develop because of topography, the presence of wetlands or bodies of water or because the land has been dedicated to some other use. Other variables measuring land supply are the likelihood of land assembly. If the parcel is
highly fragmented, the process of acquiring the land will be more difficult. If the farms are owned by a single household or firm, the expected probability of land assembly available for development will rise. We can infer that given supply of land area, indicated as a percentage of farm area normalized by whole land area, if the number of farms increases, lands are more fragmented. Therefore, using percentage of farm land and number of farms in MSAs, we could measure the impact of fragmentation and the amount of developable land to firm size. The data source are the 2002 and 2007 county censuses of agriculture. Number of counties Number of farms Farmland (acres) [Table 7 Developable land area fragmentation of land, 2002 and 2007] and

Year 2002 2007

2.5 Measuring Financial Cost and Accessibility For financial accessibility, it is hard to get the direct data, because most of the data is classified. Previous studies used the data of publicly traded companies. However, operative builders are medium or small size, and are not on the stock market. First, I used mortgage rates and mortgage costs over regions as indicators of the difference in financial markets. I also use an indirect way of finding financial accessibility, based on the assumption that national financial markets do not spread their costs across the nation. On that basis, I use quantitative amount of financial institute which can be critical to financing construction. In this case, I used number of establishment of investment banking and securities which are normalized by total number of financial establishments in each MSA (NAICS code: 523110 (investment banking), 52: financial establishment).

Each number is indexed with census variables; the list of variables is shown in attachment 12.
3.2 Effect of Disequilibrium in market on firm size. Based on literature review, demand and supply determine the equilibrium of price and stock of homebuilding. Within this equilibrium, average firm size in the market is determined. The dynamics of this process are ambiguous. However, there are variations of price and firm size in different market structures, which are useful when searching for the determinants of firm size. Based on the literature, we can assume that disequilibrium in market and regulatory influence are more critical determinants of firm size than are aggregated demand and regulation which are treated as supply costs. Therefore, we can test those disequilibrium factors' influence on firm size. First, the rent growth rate is an indicator of the market's dynamic to find an equilibrium price level. Therefore, if rents increase or decrease more rapidly than elsewhere, there is a market clearing process, responding to a demand or supply shock in the market. To test the simple relationship between rent growth rate and firm size, I picked firm size. I tested all records including null value and compared them to the correlation value based on records which did not have null value. I then test only large firm with rent growth rate which have more than five and ten employees separately. I expected a correlation with firm size. However, the correlation between firm size and rent growth rate shows little value. The rent and price of housing market lead co-linearity. I focus on price and housing price index rather than rent and rent growth, since rent level could not explain firm size. Second, regulatory variables also tested with firm size. Table 17 depicts the correlation of each variable in Wharton survey with firm size. A summary of each regulatory variable is

attached.

[Table 17 Correlation between firm size and rent] Correlation with percentage of average rent growth, Firm size: based on aggregated value of places within one Whose employees per establishment is

county

no less than 10

no less than 5

greater than 0 With all records

10.3% 2.6%

I tested all components of the sub index with firm size as I did with rent growth. The black cell means a positive correlation and white one is a negative one. The other grey cell means that there is no correlation between regulation and firm size. The ranking is shown in Table
[Table 18 Correlation test result] Variables Name Correlation 18% 17% 17% 16% 16% 13% 13% 12% 12% Local council involvement in regulation Open space index(OSI) Impact fees/exactions importance(single-family) Impact fees/exactions importance(multi family) Council opposition importance(multi family) Council opposition importance(single-family) num. of units in multi family dwelling limit multi family dwelling limit <=.5 acre minlotsize requirement

Regulations which act as cost factor for a

development

project
showed a high correlation: regulation about extraction fee and open space
endowment. Regulations affecting the approval process which is a direct cost to project show a high level of correlation. Therefore, the regulation which induces direct cost to development shows a high positive correlation, meaning that those regulations allow builders to grow, and to be more resistant to the cost effect of regulation.

% growth rate of FMR =

FMR_Q,
26 All components of regulatory index results are shown in attachment 13. 13.
6 components All of FMResults index regulatory

are shown in attachment

m Results Summary n First, although there was little variation of firm size over 20 years, the cross-sectional difference in firm size over region is dramatic. One standard deviation of mean size of establishments is an average of 97% over 287 MSA regions. Second, intensity of land use regulatory stringency requires a popular vote for zoning changes, and formal restriction such as density control on new supply and project delay times. Jurisdiction structure does not characterize highly regulated places. The results of the correlation with the firm size are below. The variation is explained with * Population and housing unit with correlation of 35-36%. * Occupation rate or vacancy rate with correlation of 36-40% *Ratio of white in racial component with negative correlation of with 23-31%. * Income level and housing value with correlation of 32-45% * Cost factor for development project: - extraction fee - open space endowment. - delayed month in approval -The variation is not explained well with * Geographic components like land area and density *Rent growth rate
*Overall regulatory stringency index *Community control and discretion *Jurisdiction structure
Chapter VI: Results and Analysis 1. Refining Somerville's model I chose the 2005 data set, because the regulatory variables are based on a 2005 survey. As indicated above, the total number of observations used to refine is 140 MSAs 27. The result is shown in Table 20. As a first step, a previous Somerville's regression model
indicated Model 0 is used as a starting point for the most recent data. Financial components, construction costs and housing values are added in Model 2, 3, and 4. [Table 19 Regression Summary table to refine previous model]

Variables Intercept In() Permit In(O Permit/Housing Unit In(_ Housing unit In()_ Farm(No) InO(_ Farm(% acre)

InMortgage cost

Model 0 -2.941 (-0.603) 0.412 (4.921)

Model 1 -1.200 (-1.293)

Model 2 -0.533 (-0.168)

Model 3 -0.284 (-0.1)

Model 4 -4.713 (-2.029)
-0.194 (-1.523) 0.253 (2.358)
0.366 (3.307) 0.448 (4.731) -0.226 (-1.648) 0.288 (2.472)
0.400 (3.440) 0.378 (2.318) -0.207 (-1.510) 0.328 (2.742)

-0.242 (-0.460)

0.293 (3.5) 0.850 (7.9) -0.372 (-1.5) 0.321 (3.3)
0.428 (3.893) 0.338 (4.819)

0.241 (2.838)

No. IV Est. per No. of Financial Est. InConstruction Cost Inmedian Housing Value
(0.679) 0.229 (1.5) 0.421 (2.010)

Adj. R2 F Obs.

0.254 7.918 140

0.268 9.951 140

0.252 15.324 140

0.318 15.0 140

0.274 5.492 140
* All models independent variables are log form of firm size (number of employment per number of establishments)
Detailed data process about number of data and data description is attached
Model I added 'demand' and 'market size' to model 0.
'Demand' represents the number
of permits that were normalized by total housing unit, meaning the ratio of newly added housing units to existing stock. Intuitively, it can be easily inferred that MSA area is correlated with homebuilders' size, shown in table 21. [Table 20 Correlation test in model 0]

Permit

Permit 1

Farm No

1 0.23

% farm

MSA area
MSA, area and number of permits
all showed high correlations of

Farm No % farm

0.49 -0.17

around

indicating
possibility of multi co-linearity. Other variables showed independency from one another. Therefore, it seems necessary to normalize the market activity with market size. I normalized 'permit' with 'total number of housing units in MSAs,' which showed the newly added stock ratio over existing market size. I measured the total housing unit number as a function of market size. 'Housing unit' is added to measure market size.

With Density Restriction Index Without Density Restriction Index

Adj. R Obs.

0.Coef. -2.20 0.47 0.75 -0.34 0.41 2.47 S.E 1.19 0.13 0.13 0.14 0.15 1.72 t -1.84 3.67 5.88 -2.38 2.76 1.44 % change of Firm size 30 24% 67% -30% 24%
0.Coef. 0.76 0.21 0.45 -0.39 0.26 3.18 S.E 1.66 0.14 0.17 0.14 0.16 1.81 t 0.46 1.49 2.70 -2.78 1.59 1.76 %change of Firm size 15% 39% -33%
Intercept In Permit/Stock In HUEST 2005 In No. of Farm In % Farm HPI g(03)
To measure the impact of the approval delay index as it pertains to the cost of development process, I grouped the data set by the approval delay index. I use first quartile as highly
delayed MSAs and forth quartile of index value as lightly delayed MSAs in processing the approval.
Table 31 reports the results, although some variables don't show statistical power, supposedly due to lack of observation for each group. The short approval delays increase the sensitivity by the market activity variable to firm size at a rate of 26%, which is greater than the 22% of long delayed MSAs. This makes sense in that less delay in the process
30 Measured by % change of firm size by one standard deviation increase of independent variable.
term of approval, makes a firm more active in their reaction to demand shock. Market size that shows a long approval index group shows more sensitivity of firm size by market size. It can be also referred that the highly regulated place with a long delay process induces costs that large firms can endure and the firm size's variation becomes wider based on some other factors, including market size.
[Table 31 Comparison of regression Results by intensity of Approval process] I s quartile of Approval delay index: longer delay 2.75 0.Coef. -0.35 0.50 0.70 -0.46 0.42 t -0.16 2.12 2.64 -1.73 1.49 % change of Firm size 22% 55% -36%

quartile of Approval

F Adj. R2 Obs.
delay index: shorter delay 3.56 0.Coef. 0.37 0.41 0.37 -0.04 0.31 t 0.21 2.23 2.13 -0.23 1.60 %change of Firm size 26% 38%
Intercept In Permit/Stock In HUEST 2005 In No. of Farm In % Farm
Chapter VII: Conclusion and policy implication 1. Conclusion The firm size as measured by employment per establishment shows wide spatial variation over regions. The variation is systematic rather than random and helps to understand homebuilders' behavior. Based on an urban spatial theory approach which assumes supply of housing is constrained to population growth in the long run, this research reports that the limited number of establishments of homebuilders in each region adjusts their size to accommodate variations in housing demand in the market. The analysis shows that the firm size at given time is affected by various factors with statistical proof from the market, land supply, and regulatory perspective. [Table 32 Results summary One standard deviation increase of Demand Market size Housing price index Land supply fragmentation based on Model 5] % increase Assuming new permit as equilibrium amount to of firm size 24%. demand for housing, market demand increases 65%. firm size by 24% from mean value. Market size of 112%31 37% region shows most significant impact on firm size. 21%

msa fipstateoJA t 13 NortheastM Midwest2lg 9I, South J J Westo-JM A ID msa name.msa 9,o 2 SPll qd 12 SCIi LZAlso-2IW2 LPAI2J1 221LAI WR2 _2 ORi. 1oJ W OS 2-2 21EIo W W221J9SRI. g2J ADI-2JWg WRLURI2OW2O-I! l fipscty 7I
name-msa 1 emp2I est!-lti size

name.msa emp est Size

[Attachment 11 Descriptive statistics of counties and MSAs while data processing]

Total counties

number in 2005 CBP data

Number of counties

whose employment number is not zero

Number of MSAs

from aggregated Counties data
(matching with other variables)
* Comparison between counties whose employment is captured and the other counties

Average of countys

Count Counties whose employment is not zero zero 1696
Housing Unit 135.91% 25,563

Land Area 1,068 1,034

Density (housing unit per land area) 371 76
Median Housing Value 129,962 90,656
* Comparison between counties within MSA and the others, among 1696 counties

Count Housing Unit

Average of county's
Land Area Density (housing unit per land area)
Median Housing Value 108,536

86,694

In MSA

Not in MSA

381 1696

76,882

10,678
Total counties number in 2005 CBP data
Number of counties whose employment number is not zero
* Among counties whose employments not zero, Comparison between counties within MSA and the other counties Average of county's Count in MSA

not in MSA

Employ -ees 281

Establish -ment 19

Size 14.8

Housing Unit 145,371

17.011

Land Area (mile 2) 1,026

Density (housing unit per land area) 398
Median Housing Value($) 131,017

116,498

41 559
[Attachment 12 Index for census data used in correlation test]
Variables GCTPH1 US25CO GCTPHI US25C1 GCTPH1 US25C2 GCTPH1 US25C3 GCTPH1 US25C4 GCTPH1 US25C5 GCTPH1 US25C6 GCTH5 US25C1 GCTH5 US25C2 GCTH5 US25C3 GCTH5 US25C4 GCTH5 US25C5 GCTH5 US25C6 GCTH5 US25C7 GCTH6 US25C2 GCTH6 US25C3 GCTH6 US25C4 GCTH6 US25C5 GCTH6 US25C6 GCTH6 US25C7 GCTP6 US25C1 GCTP14 US25CO GCTP14 US25C1 GCTP14 US25C2 GCTP14 US25C3 GCTP14 US25C4 GCTP14 US25C8 GCTH9 US25Cl GCTH9 US25C2 GCTH9 US25C3 GCTH9 US25C4 GCTH9 US25C5 GCTH9 US25C6

-4% -4% -8% 0%

-13% -1% -1% 4%
-5% 4% -4% 15% 12% 11% 7% 16% 41 0% -7% 6% 5% -11% -1% 3% 4% -7% 4% 10% 11% N/A N/A N/A N/A -4% 7% 15% 11%
-11% 0% -4% 4% 3% -4% 0% 15% 4 9% -3% 8% 12% -2% 1% 1% 21% -6% 2% 6% 8%
2% -6% -3% -4% 9% -2% 4% 17% 43 1% -7% 5% 8% 54 -5% -4% -1% 19% -9% 4% 9% 13%
1% -4% -3% -9% 1% -6% 5% 16% 44 2% -4% 6% 9% 2% 9% -7% 4% -1% 7% 3% 6%
9% 14% 10% -6% 6% -5% 13% 0% 40 -6% -5% -4% -2% -3% 1% -2% 1% 8% 10% -4% -9% 10% -5% 5% -6% -7% 3% 6% 8% -11% -6% 3% 7%

-2% 6% 2% 4%

-3% 4% 2% 9%
1% -1% 7% 10% 80 -8% -1% -8% -2%

2% -1% 6% 9%

-7% 3% -1% 1% N/A N/A N/A N/A
8% -9% -10% -12% 5% -11% -4% -5%
N/A N/A -10% -10% -4% -1% 15% 17%

N/A N/A N/A N/A

3% -6% 8% 7%

8% -3% 10% 17%

-7% -2% -9% -3%

8% -9% -10% -12%

-13% -1% 1% 5%

9% -2% 8% 10%

-8% 5% 9% 13%

-7% -3% 7% 15%

34 Each number is matched with the individual survey article, which is attached with summary statistics over MSAs.