Professional Documents
Culture Documents
Housing
Housing
(1) Introduction
For our final project in Kalamazoo Colleges BUSN-160 Winter 2016 Business Statistics
course, we Cody Colvin and Daichi Hirakawa set out to measure student
satisfaction with local off-campus housing offerings.
We wished primarily to find out how satisfied students are with their current living
arrangements, which elements of their experiences they deemed most important in
determining their overall satisfaction, and how they discovered their homes availability.
After reviewing similar studies by the University of Wisconsin-Whitewater (1) and
Longwood University (2), we crafted our questions and used a physically-distributed
survey to gather our data. We asked students, specifically, to:
A) Rate the effectiveness of property listing tools (listing websites, signs, word of mouth)
B) Indicate (through a 1-5 rating system) variables which most significantly contributed
to their overall satisfaction
C) Rate in several categories (1-5 scale) the service quality they were receiving from
their landlords, and
D) Include financial information, including monthly rent and utility costs.
Throughout our study, we communicated closely with a client, Mike Kiewiet. A prolific
local property owner, Mike owns and rents out 11 properties in our surveyed
neighborhood, and is actively involved in the tenancy of an additional six. We reached
out to him under the assumption that he could most effectively use our data to optimize
future housing offerings for local college students, and in doing so, run his business
more efficiently and profitably. His extensive connections with other area landlords also
assured us that he could disseminate any useful findings to the rest of the local real
estate community.
Throughout our entire data-gathering process, we made the conscious decision to exclude data
of non-student residents. Though such information mightve been useful to our client, we
decided to remain true to our original intention in helping to optimize the local student housing
market.
After visiting 28 houses (we immediately subtracted 3 houses which we assumed were not
student-rented), we removed from our sample those which we discovered through direct contact
were also not inhabited by students.
Of identified student houses from our final sample, we received a response rate of 18/22,
around 82%. Specifically, we received received responses from 40 students living in those 18
houses. Our data was collected over the course of two excursions on a Sunday afternoon, one
early in the day, and one later at night. We attempted to contact residents from the entire
sample on the first visit, and tried once more that night to contact residences which hadnt
answered previously. To ensure the greatest level of randomness, we did not replace
unresponsive residences, and we approached only those houses selected by our random
number generator.
Ultimately, we consider our final sample an accurate representation of our population. Our
respondents informally affirmed our prior knowledge that Mike Kiewiet owned a substantial
section of those three blocks; thus, we knew that the data sample wed provide to Mike
consisted largely of the reviews and preferences of his own tenants.
The following table lists the average rating of each variable, the standard deviation for each
rating, and each variables median and mode. The graph which follows it provides a visual
representation for the means of each variable. We consider these statistics highly useful to local
landlords, especially our client; knowing exactly what students want allows managers to
optimize their offerings for maximum customer satisfaction.
AMENITY
Average
Importance
Rating (1-5)
Standard
Deviation
Median
Mode
4.675
0.83
Proximity to
campus
4.625
0.63
4.575
0.75
Familiarity with
roommates
4.525
0.82
Affordability of
monthly cost
4.425
0.78
4.35
0.83
Landlord fairness
VALUE
0.5
1.5
2.5
3.5
4.5
AMENITY
AMENITIES
6
AMENITY
Timeliness and
quality of
maintenance
Average
Importance
Rating (1-5)
Standard
Deviation
Median
Mode
4.15
0.98
3.975
1.05
3.95
0.9
3.875
0.91
Readily available
parking nearby
3.775
1.29
Affordability of
down payment
3.625
1.27
Size of home
overall
3.525
1.04
3.5
Previously
installed air
conditioning
3.25
1.5
Proximity to
restaurants /
nightlife
3.025
1.1
2.9
1.41
2.5
Presence of
outdoor patio/
porch
2.775
1.54
2.5
Pets permitted
2.375
1.25
Size of yard
2.225
1.23
Having furniture
provided by
landlord
1.775
0.97
1.5
Landlord
responsiveness
Safety of
neighborhood
LANDLORD
RATING
CATEGORY
Kindness
Average Rating
Confidence
Interval (95%)
Standard
Deviation (s)
Standard Error
(s/n)=(s/6.3246)
0.875595
0.138443
Fairness
0.846940
0.133912
Maintenance
0.971056
0.153536
1.042679
0.164861
Privacy
The average ratings listed above signify the mean rating for each category, and the subsequent
confidence intervals were determined by calculating the standard error of each variable and
multiplying it by the t-value corresponding with 95% significance and 39 degrees of freedom,
2.023. Our confidence intervals suggest that we can be 95% certain that the true ratings for the
entire population of students in our target neighborhoods lie within those ranges.
Ultimately, students seem exceptionally satisfied with the service theyve received from their
property managers. We also believe that landlord service satisfaction could play an
exceptionally important role in determining overall satisfaction, discussed later in our Multiple
Regression section.
Average rating
(1-5)
Overall
Satisfaction
Confidence
interval (95%)
Standard
Deviation (s)
Standard Error
(s/n)=(s/6.3246)
0.6
0.095
Similar observations follow for overall satisfaction. Students are highly satisfied, and we can be
95% confident that the true population rating for satisfaction would lie somewhere between 4.31
to 4.69.
Website
Avg.
Effectiveness
(1-5)
Conf. Interval
(95%)
Standard
Deviation
Standard Error
(s / n) (s/6.3246)
Zillow
1.679
0.265
Trulia
1.471
0.233
craigslist
1.285
0.203
apartments.com
1.328
0.210
wmuhomes.com
1.079
0.171
showmetherent
0.84
0.133
hotpads
0.56
0.089
Some of our most actionable information is our discovery of listing sites effectiveness. The
average effectiveness again represents the mean rating of the sites importance (1-5, the latter
signifying greatest importance), as students were asked to indicate how important each site was
in helping them find their off-campus home. The confidence intervals were formed by adding
and subtracting the product of the standard error (listed on the far right) and the t value for 39
degrees of freedom at 95% confidence, 2.023.
As listed above, Zillow is by far the most effective site for disseminating listings information.
Based on this information, we suggested that our client focus advertising efforts on Zillow; he
then informed us that he uses a service, rentlinx.com, which posts listings on a plethora of sites,
including Zillow and Trulia.
With the comparatively high use of Zillow, we thus suggest that landlords near Kalamazoo
College provide detailed information on the popular site. A service which posts on multiple
websites is further advised, as our data suggests a tendency for students to rely primarily on
one or two sites in their property search.
UTILITY
Provided by landlord
Confidence Inverval
(95%)
Standard Error
Electricity
6.6%
Gas/heat
6.6%
Water
Garbage
Snow
7.25%
6.6%
6.3%
Cable
0% None
None
Internet
0% None
None
We also believed it beneficial for our client to know, with as much accuracy as possible, which
utilities were typically included in renters fixed monthly costs. The sample proportions for each
variable are listed on the left, with confidence intervals (and corresponding standard errors)
listed to the right.
Each standard error and confidence interval is calculated at the 95% level of significance; thus,
we can say with as much certainty that the true proportion for our entire population lies within
those specified ranges. Acknowledging a 5% chance of error, we found garbage and snow
removal to be the only majority landlord-provided utilities in the area. Our client acknowledged
that the primary reason he pays for garbage removal is due to the observed tendency of
students to let it accrue to unwieldy amounts. Thus, this is unsurprising.
We did not, however, expect such results for the inclusion of cable and internet in monthly cost.
Since both of these variables are commonly flat-rate amenities, it might make sense for
landlords to factor it into rental costs. At the very least, these results inform our client that his
lack of cable and internet provision is not likely unique.
It is worth noting, however, that desired cable and internet packages likely differ from tenant to
tenant. Purchasing a cable subscription for a tenant who dislikes television, and expecting for
them to pay for it in their monthly cost, would likely lower that tenants overall satisfaction.
Furthermore, the variability of cost by package, rather than use, makes cable and internet
similar to utilities such as gas or water, in that its cost varies with its use.
10
p =.55
Using a one-tailed test (due to our interest in only one critical value direction), we can then
conclude with 95% confidence that our true population proportion the percentage of K
students in our surveyed neighborhoods which would rate affordability of cost as most
importantis less than .76.
We cannot, however, claim that K students care less about cost than SDSU students. Sample
statistics are inherently imperfect, and in addition, we did not test the significance of SDSUs
proportion. It is indeed possible that, by these metrics, K students actually do value cost less,
but such a comparison is impossible to make accurately with our resources and sample size.
11
H0: 1
HA: < 1
x = 1.375
critical region: reject any sample mean less than 1-tse; reject
any sample mean less than .667
DO NOT REJECT NULL HYPOTHESIS
Thus, by not rejecting our null hypothesis, we find with 95% confidence that our true population
mean is not less than 1. The difference between our sample mean and an estimate less than
one cannot be explained by random chance. Thus, our data does not conform with UWWs
findings of their own students. Why our students seem to visit more properties is difficult to
guess; UWWs housing offerings might be more homogenous, forcing students to visit fewer
properties before getting a feel for their options. Ultimately, its difficult to say.
12
We found a mean monthly rental cost (not including utilities) of $366 with a standard deviation of
$96, with both numbers rounded up to nearest dollar. To determine whether this cost is
significantly different from the billed cost of standard double room at Kalamazoo College (4), we
used hypothesis testing.
According to the Colleges website, a standard double costs $4335 per annual school year (9
months); thus, the monthly rate for such a room computes to $481. Our null hypothesis states
that the difference between our rental average ($366) and the colleges monthly rate ($481) can
be explained by random chance. The alternative hypothesis states that there is, indeed, a
statistically significant difference at the 95% level of confidence.
signifies the true population average rent for K students in our surveyed neighborhoods. The
form of our question specifically, whether the sample mean of off-campus rent is significantly
less than on-campus room cost allows us to use a one-tailed t-test, as the form of our
question only inquires as to whether off-campus cost is less than the tested idea.
H0: = 482
x = 366
Thus, with 95% confidence we can reject the idea that our population mean might equal $482;
there is a statistically significant difference between our sample mean and that number. It is
important to remember, however, that we did not factor our monthly utility average into this test.
In any case, it seems much cheaper to live off-campus than on-campus, even including utility
cost (our average was $40). Thus, we can say with at least some confidence that off-campus
housing can be a much more affordable choice than remaining in K Colleges dorms.
13
Looking at our data, we found that the average frequency of property showings for males was
1.7, with a standard deviation of 1.74; for women, it was 1.1, the standard deviation 1.25. Using
hypothesis testing, we sought to answer the following question: did men in our population make
use of more property showings than did women?
We formed a null hypothesis stating that there was no difference, and that the numerical
discrepancy can be attributed to random chance. Our alternative hypothesis, which we expected
to be true, states the opposite, that there is indeed a statistically significant difference between
the two population means.
s1=1.74, n1=20;
s2=1.25, n2=20
x 1 - x 2 = 0.6
Thus, by not rejecting the hypothesis, we know (with 95% confidence) that the difference
between the two sample means might be a result of random chance; our guess turned out to be
incorrect. Thus, we cannot accurately say that our population of males attend more official
property showings than our population of females.
14
(Test 2) Are males more satisfied with off-campus housing than females?
After surveying 20 males and 20 females, our overall satisfaction means for males (4.55,
standard deviation of .605) and females (4.47, standard deviation of .612) looked extremely
similar. We wished to prove that there was no statistically significant difference between the
variables, so we formed our null hypothesis to state that the true population means are equal.
The alternative hypothesis states that there is, indeed, a statistically significant difference
between the true means, not explainable by random chance.
H0: (-) = 0
x 1 - x 2 = .08
Thus, we are able to confirm our theory, and we do not reject the null hypothesis. At the 95%
confidence level, there is no statistically significant difference between the population means of
males and females in overall satisfaction.
15
Similarly to the previous test, our data indicates close means for satisfaction between juniors
and seniors (4.58 and 4.6, respectively). However, we wished to further prove this by running a
hypothesis test. Our null hypothesis states that there is no statistically significant difference
between the true population means for this variable; our alternative hypothesis states that
seniors, whose mean overall satisfaction in our sample is slightly higher than juniors, have a
higher overall average satisfaction rate (in the population) than do juniors. These results
represent the responses of 12 juniors, and 25 seniors, with standard deviations of .669 (juniors)
and .577 (seniors).
H0: (-) = 0
x 1 - x 2 = .02
By not rejecting the null hypothesis, our estimation that seniors are not more satisfied than
juniors seems correct at the 95% level of confidence. One might think that juniors and seniors
would have different expectations of off-campus housing; indeed, its easy to assume that
seniors, likely having already spent a year off-campus, might have different expectations for
service. At the very least, this test suggests to us that years spent off-campus would probably
not serve as a suitable variable for a regression of overall satisfaction.
16
3e: Multiple Regression
17
(.93568)
(.14658)
(.13357)
(.15414)
Our predictions regarding the direction of each correlation were correct. Importance of friends/
word of mouth is indeed negative, with importance of sites and importance of signs both
producing positive correlations.
The equation suggests that for every point added to the ratings of importance for each variable
(on a 1-5 scale), the dependent variable increases by that increased variables coefficient. This
means that a one-point increase in the importance of property signs corresponds to a .29
increase in the number of showings attended.
After calculating the equation, we used a shortcut to determine the statistical significance of the
equations dependent variables. By looking at the p-levels, and seeing that none of the variables
possess a p-value lower than .05 (at 95% confidence), we were able to conclude that none of
them can be considered statistically significant at that confidence level. In other words, their
relation to the independent variable could merely be the result of random chance.
The low value of R-square indicates that many influential variables outside of this equation help
to determine the number of official showings students attend. Due to the simplicity of the
equation and the scale of our study in general, such results were expected.
Based on this equation, installing more property signs seems to have the best chance of
securing more official showings. It is important to note, however, that this does not necessarily
correspond with an increased volume of prospective renters, and that this equation is merely a
simple attempt with very limited data to understand a complex market variable.
18
3e: Multiple Regression
When we were planning the course of this study, our initial intention was to craft a regression
equation for overall student satisfaction with off-campus housing. While we found that our
sample is, overall, extremely satisfied with a mean satisfaction rating (from 1-5) of 4.5, a
standard deviation of .6, and a mode and median of 5 we encountered significant difficulty in
creating a reliable regression equation for the variable.
The dependent variable is students 1-5 rating of overall satisfaction. The independent variables
are as follows:
Number of roommates
Rated landlord kindness
Rated landlord fairness
Rated landlord commitment to timely maintenance
Rated landlord privacy
It seemed logical that students perceptions of their landlords would factor heavily into their
overall satisfaction. However, we did not anticipate a negative correlation coefficient for any of
the landlord ratings; we figured that, as ratings went up, so too would overall satisfaction.
In reality, it seems that the correlation coefficients for the independent variables we selected
were determined largely by random chance. We can tell by the size of our r-square; a low rsquare value means that many important, unidentified variables lie in the background.
Indeed, unlike our previous regression equation, this one seems quite illogical. Only through
randomly testing correlations were we able to determine that the number of roommates a
student has corresponds quite heavily to overall satisfaction, and our assumption of a positive
correlation was correct. In fact, using the p-level test rejecting the idea that there is no
significant relation at 95% confidence if the p-value is below .05 tells us that the roommate
variable is the only one which cannot be explained by random chance, as every other variable
(not including the intercept, which for us is largely inconsequential) sits well above a p-value
of .05.
19
(.576)
(.049)
(.313)
(.327)
(.1)
Its possible that the variables interfere with each other through multicollinearity, obscuring the
regression due to interactions between independent variables. Its also possible that more
roommates truly leads to greater student satisfaction. Our statistical analysis here is limited; like
our previous equation, we are attempting to explain an extremely complex variable this one
psychological, as much as economic with limited data and only a few variables.
Ultimately, we are less confident in this regression equation than that of property showings, and
understandably so: this dependent variable is far more nebulous, and perhaps well beyond our
capabilities. Still, the discovery that the number roommates a student has is somehow
correlated with overall satisfaction if that is, indeed, what weve found here is curious, and
might merit more investigation.
20
4. Conclusions
Ultimately, our statistical analysis informs us, and our client, that students in the Stone,
Eldred and Stanwood neighborhoods are highly satisfied. Our analysis also provides an
exceptionally useful list of amenities, and a measure numerical importance of each
variable.
We also found that, even when adding the average cost of utilities ($40) to rent, living
off-campus is significantly cheaper than living in a standard double room at Kalamazoo
College. This information was released informally to the student body through
Facebook, and was met with exceptional appreciation from students doing the math for
themselves.
We did not come up with a significant, breakthrough regression equation for either
overall satisfaction or frequency of property showings. This was not our intention, and
we did not expect the bulk of our analysiss value to come from these sections. We
simply do not have the resources nor the expertise to be especially useful here.
Had we more time and/or resources, we probably would have expanded our
geographical sampling range, as well as our sample size. We wouldve made more of
an effort to include Western Michigan University students, of whom only a few are
represented in our results.
Ultimately, however, the analysis we provided to our client will help him focus on offering
amenities students find important, and if we have at all positively impacted the local
housing market or, for that matter, the lives of local students then our work will have
been worth it.
21
References
(1)
(2)
(3)
(*)
(*)
Numbered citations refer to in-text citations. (*) denotes a consulted source for numerical reference.