The Gender Gap in Price Negotiations: Evidence from New Car Sales

Transcription

The Gender Gap in Price Negotiations: Evidence from New Car Sales
The Gender Gap in Price Negotiations:
Evidence from New Car Sales∗
Ambarish Chandra
Sumeet Gulati
James Sallee
March 12, 2014
Abstract
The sharp increase in the educational attainment and labor force participation of
women has constituted one of the biggest societal changes in North America over the
past few decades. Yet, economic research into the consequences of this phenomenon has
been restricted almost exclusively to the labor market. We investigate whether women’s
performance in price negotiations has improved relative to that of men. Using detailed
transaction level data from the automobile industry, we find that the age premium rises
more steeply for women than for men. In particular, young women negotiate better
than men of the same age, while older women perform worse. We evaluate several
possible explanations for our results, and conclude that higher levels of education and
work experience in the younger cohorts are the most likely cause of women closing the
gender gap in this industry.
Keywords: Gender; Age; Automobiles; Negotiations; Bargaining. JEL Codes: J14,
J16, L62.
∗
Chandra: Department of Management, University of Toronto (on leave) and Department of Economics,
University at Albany, [email protected]; Gulati: Food and Resource Economics, University of
British Columbia, [email protected]; Sallee: Harris School of Public Policy, University of Chicago,
[email protected]. We are grateful for helpful suggestions from Kathy Baylis, Jim Brander, Don Fullerton,
Keith Head, John Jones, Nisha Malhotra, Carol McAusland, and Torrey Shanks.
1
1
Introduction
The last several decades have seen dramatic increases in the educational attainment and
labor force participation of women. In North America, women now make up about 45% of
the labor force, and almost 60% of college graduates, a stark change from the 1970s.1 This
phenomenon has accompanied huge societal changes, on the rate and age of marriage, and in
fertility rates.2 Profound economic impacts, on female wages and labor outcomes have also
emerged. The gender wage gap has narrowed, women are more likely to be top executives,
and male and female MBAs now start their careers with almost identical incomes.3
While assessing the impact of these transformative changes, economists typically evaluate
the labor market, but remarkably few others. In particular, we are unaware of studies
evaluating changes in other markets where negotiation matters.4 Does increasing educational
attainment, and labor force participation allow female customers to negotiate better market
outcomes?
Personal interaction and negotiation influence the price or quality in many markets.
We negotiate with service providers, financial institutions, and importantly, for houses and
automobiles. In this paper, we ask whether older women today, who on average have lower
education levels and had a lower likelihood of having been employed when in their 20s and
30s, fare worse in price negotiations than younger women. We also compare price negotiations
across female cohorts relative to men of the same age, treating men as a useful control group
not undergoing the same transformative experiences.
The ideal market for such a study has data on negotiated prices and consumer demographics, and transacts identical goods.5 However, prices are generally not open to negotiation
in many everyday markets. Conversely, it is difficult to obtain price data in markets with
negotiated prices since these usually occur in private or informal settings; for example, when
bargaining for loans with financial institutions, selling goods such as used cars via Craigslist,
or hiring contractors to perform residential services.6
We look to the new car market, which we believe offers the ideal setting for our study for
a number of reasons. First, final prices in this market are commonly negotiated and there
can be differences of hundreds of dollars in the amounts that different consumers pay for the
same new car. Second, this market involves the sale of many identical goods: knowing the
make, model and trim of a vehicle pins down almost exactly the type of good that different
1
See Table 299 in the Statistical Abstract of the United States, Bureau of the Census; Table 2 in Women
in the Labor Force, Bureau of Labor Statistics; and Turcotte (2011)
2
See Goldin et al. (2006)
3
See, respectively, Blau and Kahn (2007) and O’Neill (2003); Bertrand and Hallock (2001); and Bertrand
et al. (2010)
4
There is a large literature on gender differences in wage negotiations; see, for example, Babcock and
Laschever (2003). However, these are studies of static differences, and no prior study has examined whether
these differences have changed over time. Other fields, particularly Psychology, have investigated differences
in price negotiations, though also in a static setting and primarily in laboratory settings; see Stuhlmacher
and Walters (1999) for a survey.
5
Alternatively, it has data on varying characteristics to control for systematic differences in the choice of
goods by different types of consumers.
6
The housing market is a good possibility, but it is rare to find housing transactions that involve identical
units — a sample of housing transactions will always involve considerable heterogeneity in characteristics
that are difficult to control for precisely.
2
consumers buy. Other differences in transactions, such as the location of car dealers and
the timing of the transaction, can easily be controlled for. Finally, we have data on two
key elements of new car transactions: the final price paid by consumers, and the dealer’s
invoice price, which represents the opportunity cost to the dealer of selling the car. We can
construct precise measures of the dealer’s profit margin—rare in almost any industry—and
examine whether these margins are systematically different for different types of consumers.
Following Busse et al. (2013), we estimate equations for transaction prices and dealer
margins controlling for a range of demand and supply covariates. Our data come from a
major marketing firm, and are drawn from a very large set of new car transactions in Canada
between 2004 and 2009.7 Exploiting the large size of our sample, we study the variation
in dealer margins within model, year, province and trim combinations, and across several
gender and age categories. Our estimating strategy minimizes the potential impact of other
influences: whether the vehicle was leased or financed; whether it was purchased at the end
of the month or year, when dealers face incentives to increase sales; whether certain dealers
were more likely to offer discounts in a manner correlated with customer demographics;
whether the vehicle was in greater demand—measured by the average time the particular
model stayed on dealers lots—and whether there was a trade-in vehicle associated with the
transaction.
We find an important interaction between age and gender in the market for new cars.
Dealer margins are higher among older customers, even after controlling for all observable
aspects of the transaction. Further, margins rise faster with age for women than for men.
On average, women in the youngest age category generate $200 less in dealer margins than
women in the oldest category. Younger women also generate $100 less in dealer margins
than men of the same age, while older women generate margins that are $50 higher than
their male counterparts. All of these differences are highly significant and hold across many
different subsamples and specifications.
What explains these disparities? We consider two broad possibilities. Either the same
consumers fare progressively worse as they age (which we term an ‘ongoing’ effect), or our
findings are particular to the cohorts observed (a ‘cohort’ effect). Either way, our preferred
hypothesis explains why women do worse in the new car market as they age when compared
to men. As part of the ongoing effect, we consider differences in search and negotiation costs
across age. These can derive from differences in: access to the Internet, a perceived value of
time, or aversion to change.8 We also consider gender based differences from the literature
on ‘sensation seeking.’9 Despite our broad reach, we cannot explain why margins rise faster
with age for women than men.
We then consider the cohort effect. This suggests that younger women negotiate better
due to their superior educational attainment and labor market outcomes. By contrast,
7
We believe our findings extend to the US market as well. The automobile industries in Canada and the
US are extremely integrated and very similar in the process of customer negotiation for new cars. Moreover,
Canada has undergone demographic changes very similar to those of the US in terms of increases in women’s
education and labor force participation. In fact, the gender gap in education was eliminated a few years
earlier in Canada than in the US, and currently women in Canada perform better, relative to men, in Canada
than in the US. We present detailed comparisons using OECD data in Table 19 in the Appendix.
8
See Morton et al. (2001), Goldfarb and Prince (2008), and Lambert-Pandraud et al. (2005)
9
See Zuckerman (1979) p. 10.
3
women above the age of 65, who generate the highest dealer margins, are less likely to
have participated in the labor force when they were younger, or to have obtained a college
degree, than women under 25, who generate the lowest margins. These differences can
potentially cause older women to have lower information in the new car market and possibly
also negotiate with less confidence. If this hypothesis is correct, it implies that today’s cohort
of young women are unlikely to do worse than their male counterparts as they age.
We further explore this hypothesis by evaluating differences in this relationship across
Canadian provinces: if cohort effects explain our observed gender gap, these effects should
vary across provinces where female employment and education outcomes differ. We consider
Canada’s two most populous provinces, Quebec, and Ontario. Women in their 30s in Quebec
are 5 to 6 percentage points more likely to be employed, relative to men, than in Ontario.
Similar differences exist in education across the two provinces. Consistent with this hypothesis, we find that women in Quebec perform better in the new car market, relative to men,
than in Ontario. This leads us to consider the cohort effect as our preferred explanation.10
Previous research into the automobile industry has examined the role of gender and
other demographics in price negotiations. Famously, Ayres and Siegelman (1995) used an
audit study and found that women and minorities are disadvantaged in the new car market,
as both groups are offered higher initial prices by dealers as well as negotiate higher final
prices. Since then, a number of studies have re-examined the issue of gender differences in
the new car market, including Goldberg (1996) and Harless and Hoffer (2002). Both those
studies cast doubt on the finding that women pay more for new cars. More recently, Morton
et al. (2003) show that, while minority customers pay a higher price than others, this can be
explained by their lower access to search and referral services. Similarly, Busse et al. (2013)
find that women are quoted higher vehicle repair prices than men when callers signal that
they are uninformed about prices, but these differences disappear when callers mention an
expected price for the repair.
Importantly, no previous study has examined the interaction of age and gender in the
new car market, especially in the context of possibly different behavior exhibited by different
cohorts.11 Our study does so explicitly, and also makes other contributions to the existing
literature. First, we examine more than half a million transactions in the new car market, a
far larger sample than was available in the earlier studies described above. Second, we have
detailed data on the characteristics of each vehicle sold, as well as on both final transaction
prices and dealers’ invoice prices. This allows us to calculate dealer margins, rather than
relying only on the transaction price which includes many unobserved components. Third,
we have demographic data on the actual consumers in all of our transactions. This allows
us to directly examine how age and gender relate to dealer margins.
In Section 2 we present the data used in our study. In Section 3 we present the empirical
framework. Section 4 contains our main results along with various checks for the robustness
10
While we find suggestive evidence that differences in negotiating outcomes are likely driven by labor
market and education changes, we do not conclusively prove this. Our data are a snapshot of the new car
market, and we do not follow cohorts over time. The ideal dataset would involve multiple repetitions of our
data over a very long period of time. But such data do not currently exist.
11
Harless and Hoffer (2002) do examine customer age, finding that older consumers generally pay more on
average, but do not disaggregate the relative performance of each gender across age cohorts. Langer (2011)
examines the interaction between gender and marital status, but does not study cohort effects either.
4
of our findings. Section 5 discusses various explanations for our results. We provide a brief
conclusion in Section 6.
2
Data
We use data provided by automobile dealers to a major market research firm. These data
include more than 250 key observations for each vehicle transaction including: a) vehicle
characteristics: vehicle trim, number of doors, engine type, transmission, cost of the vehicle
including factory and dealer installed accessories and suggested retail price; b) transaction
characteristics: transacted price, rebates offered, how long the vehicle was on the dealer’s lot,
and whether the vehicle was financed, leased or a cash purchase; c) trade-in characteristics:
the price of the trade-in vehicle, and its under- or over- valuation; d) customer demographics:
gender, age, province of residence, and province of purchase. The selection of dealers is not
random, as dealers must agree to be included in the database. However, it is the most
comprehensive dataset on new vehicle purchases that we are aware of. Our sample of the
data is approximately 20% of new vehicle sales in Canada from May 1st, 2004 to April 15th,
2009, a total of 1,137,573 transactions.
2.1
Sample Selection
To examine the representativeness of our sample we compare our dataset with aggregate sales
data provided by Desrosiers Automotive Consultants. Table 1 presents the distribution of
transactions by major manufacturer. The first two columns are the number of transactions
and percentage of overall transactions for the year 2006 in our sample. The third column
is the percentage of sales by manufacturer from data compiled by Desrosiers Automotive
Consultants for the same year, which captures all sales in Canada. Our sample somewhat over-represents sales by GM, Toyota, Nissan, and VW, and therefore under-represents
Chrysler, Ford, Honda, and other manufacturers. However, the differences are not large.
Table 2 presents the distribution of transactions across provinces for the year 2006. The
first two columns are the number of transactions and percentage of total transactions in our
data. The third column is the sales share calculated from data provided by Desrosiers, and
the fourth column has the percentage of population for each province.12 Our sample clearly
over-represents Alberta, and to a certain extent Quebec, at the expense of slight underrepresentation of the other provinces. These tables most likely illustrate dealer selection,
although despite this, we are struck by how representative our sample in fact is. Nevertheless, none of our results will draw conclusions across provinces and manufacturers. As we
discuss later, our empirical strategy focuses on effects within provinces and car models, and
some results will also restrict attention to effects estimated within dealers.
We drop observations where the gender or age of the consumer is not available. We also
drop observations with very small and very large dealer margins—the bottom and top 1%
of the data. These correspond to unusual models, typically with very high purchase prices,
12
Source: Statistics Canada, Canadian Socioeconomic Database, table 051-0001.
5
Manufacturer
GM
Toyota
Chryl
Ford
Honda
Nissan
Korea
Mazda
VW
Japan
Europe
BMW
Total
No.
Column
%
53458
31859
22703
20934
18732
9603
7232
9001
6689
5083
2633
2174
190101
28.1
16.8
11.9
11.0
9.9
5.1
3.8
4.7
3.5
2.7
1.4
1.1
100.0
Aggregate
Sales %
26.0
12.2
13.7
14.2
10.3
4.1
6.2
5.0
2.7
2.4
1.9
1.2
100.0
Source: Authors’ Calculations
Table 1: Sales Across Manufacturers (2006).
No.
Column
%
Saleshare
%
Popshare
%
65453
51728
42660
19156
2501
3310
1904
2040
710
487
152
190101
34.4
27.2
22.4
10.1
1.3
1.7
1.0
1.1
0.4
0.3
0.1
100.0
37.5
24.7
14.9
11.7
1.5
2.7
2.9
2.4
NA
2.1
NA
100.0
38.9
23.4
10.5
13.0
1.6
3.6
2.9
3.0
0.1
2.3
0.1
100.0
Province
Ontario (ON)
Quebec (QC)
Alberta (AB)
British Columbia (BC)
Newfoundland and Labrador (NL)
Manitoba (MB)
Nova Scotia (NS)
Saskatchewan (SK)
NorthWest Territories (NT)
New Brunswick (NB)
Yukon (YT)
Total
Source: Authors’ Calculations, Desrosiers Data, STATCAN CANSIM Table 051-0001.
Table 2: Sales Across Provinces (2006).
6
Segment
Compact (37%)
Full (0%)
Luxury (4%)
Midsize (13%)
Pickup (13%)
SUV (20%)
Sports (1%)
Van (9%)
Total (100%)
Mean
Median
Male
Cust
Female
Cust
Cust
Price
Dealer
Marg
Turn
Days
Cust
Age
Cust
Price
Dealer
Marg
Turn
Days
Cust
Age
46.7%
75.8%
68.9%
62.0%
83.0%
59.9%
67.6%
67.1%
59.6%
53.3%
24.2%
31.1%
38.0%
17.0%
40.1%
32.4%
32.9%
40.4%
$19,936
$35,285
$47,712
$28,617
$39,315
$35,555
$42,690
$28,423
$29,255
$1,314
$1,204
$2,701
$1,647
$2,209
$1,943
$3,075
$1,346
$1,705
44.8
55.3
53.4
58.9
61.0
49.8
58.9
71.3
53.0
44.5
67.8
50.4
51.3
43.4
45.6
43.1
47.8
46.0
$19,529
$35,215
$44,968
$27,871
$38,604
$32,219
$35,041
$27,800
$26,373
$1,238
$1,224
$2,550
$1,555
$1,865
$1,702
$2,240
$1,361
$1,463
14
8
19
20
26
16
18
36
18
44
70
50
51
43
45
44
45
46
Source: Authors’ Calculations
Table 3: Statistics across Segments.
which can potentially skew the results.13 This leaves us with about 510,000 observations
(from over a million earlier). All summary statistics presented in this data description, as
well as our later regressions, are from this smaller subset.
2.2
Summarizing Our Data
In Table 3 we present selected summary statistics across vehicle segments. The bottom row
summarizes the entire dataset. The median vehicle sold for approximately $26,373, generated
$1463 in margin for the dealer, was on the lot for 18 days, and was bought by a 46 year old
male customer.14 Luxury vehicles have the highest values. Compact cars and SUVs are sold
quicker (stay on the dealers lot for fewer days) than other segments.
In Table 4 we present a cross-tabulation of data on consumers’ ages and gender. We
divide age into 11 categories of 5-year intervals. Approximately, 59% of all transactions
have male customers. Women outnumber men in the under 25 age category, but not in any
other category. We observe an interesting pattern of ‘missing women’ (see columns titled
Row % in Table 4): from comprising almost 52% of transactions among under 25 year-olds,
women comprise just 32% of all transactions consumers over 70. There are a few possible
13
This entails dropping observations where the dealer margin was greater than $7300, or where the dealer
lost more than $2200 on the sale.
14
Vehicle price for each transaction is the price listed on contract before taxes/title fees/insurance. In
our datset, this is “[T]he price that the customer pays for the vehicle and for factory and dealer installed
accessories and options contracted for at the time of sale, adjusted by the trade-in allowance, but not adjusted
for the customer cash rebate amount, if any. This price is not the Manufacturer’s Suggested Retail Price
(MSRP).” Note that the vehicle price adjusts for manufacturer to dealer rebates, for under or over allowance
of trade-in. Further, in this variable, we also include customer cash rebates, and the price for extended
warranties. The vehicle cost is the dealer’s cost for the vehicle. Defined in our dataset at “[T]he retailer’s
‘net’ or ‘dead’ cost for the vehicle and for both factory and dealer installed accessories contracted for at the
time of sale.” This includes transportation cost, and also includes a measure of installed accessories. Dealer
margin is the difference between vehicle price and dealer cost.
7
Gender
Age Category
Age Under 25
Age 25-30
Age 30-35
Age 35-40
Age 40-45
Age 45-50
Age 50-55
Age 55-60
Age 60-65
Age 65-70
Age Over 70
Total
No.
Male
Col
%
Row
%
No.
Female
Col
%
Row
%
No.
Total
Col
%
Row
%
15301
22801
28251
31889
36553
38623
35561
29708
21921
15341
22554
298503
5.1
7.6
9.5
10.7
12.2
12.9
11.9
10.0
7.3
5.1
7.6
100.0
47.8
52.3
58.0
60.0
58.1
57.4
58.0
59.8
62.4
64.3
68.1
58.4
16716
20776
20493
21270
26379
28631
25789
19976
13231
8531
10571
212363
7.9
9.8
9.6
10.0
12.4
13.5
12.1
9.4
6.2
4.0
5.0
100.0
52.2
47.7
42.0
40.0
41.9
42.6
42.0
40.2
37.6
35.7
31.9
41.6
32017
43577
48744
53159
62932
67254
61350
49684
35152
23872
33125
510866
6.3
8.5
9.5
10.4
12.3
13.2
12.0
9.7
6.9
4.7
6.5
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
Source: Authors’ Calculations
Table 4: Sales by Age Category
explanations. First, women who purchase vehicles with male partners may be less likely to
be listed as the owner of the vehicle. If so, and if the likelihood of women being attached
increases with age, they will correspondingly also disappear with age. Alternatively, women
from older age cohorts had lower employment rates than younger women (see Table 17 in
the Appendix), and likely also earned smaller salaries while employed, leading to a lower
propensity to purchase new vehicles.
In Figure 1 we illustrate how the mean transacted price, vehicle cost and dealer margin
varies with age in our data. Both mean price and cost peak for customers aged 39 years, at
$30,306, and $28,900 respectively. Like the average price and cost, dealer margin rises with
age and then declines after a peak of $1769 at age 50.
Table 5 breaks down vehicle prices and dealer margins by segments and gender. Overall,
men purchase more expensive vehicles, with an average price of $31,075 versus $26,589 for
women. Male customers also generate higher margins for their dealers at $1,784 versus $1,590
for female customers. This difference carries through within vehicle segments as well.
3
Estimating Strategy
Our goal is to examine whether there are systematic differences in the performance of various
demographic groups when negotiating prices in the new car market. If we had complete
data on all of the characteristics involved in new car transactions, we could simply examine
how final transaction prices vary by demographics, while controlling for these observable
characteristics. However, despite the high level of detail in our data, there remain some
characteristics that we do not observe. For example, while we know the make, model,
model-year and trim of the vehicle sold, as well as the date of the transaction, we do not have
8
Figure 1: Prices, Vehicle Cost and Dealer Margin by Age
30000
Cost
1700
Dealer Margin ($)
Price and Vehicle Cost ($)
Price
28000
26000
1600
1500
24000
20
40
60
80
20
Age
40
60
80
Age
direct information on certain options and after-market purchases, such as service contracts
and accessories, which may affect the transaction price.
However, the information on these options and accessories is embedded in our data on
the vehicle’s invoice price, which the dealer records and reports to our data provider. For
this reason, examining the dealer’s margin as opposed to the transaction price, is sufficient
for our purposes. If certain demographic groups generate systematically different dealer
margins, holding constant the attributes of the purchased vehicle and other features of the
transaction, then we can confidently attribute these differences to the negotiating process.
We keep in mind two goals. First, to examine dealer margins within as uniform a product
as possible. Our very large sample size is an incredible asset for this. There are many high
selling vehicle models, with over 1,000 transactions for each model-year combination (some
with over 5,000 transactions). These popular models (such as the Toyota Corolla and Honda
Civic) draw customers from a wide distribution of gender and age, as shown in Figure 2.
Our data allows us to identify the within model, year, province and trim variations in vehicle
price and dealer profits across several gender and age categories.
Second, to minimize the potential bias from omitted variables. Recent research indicates
that besides model and dealer characteristics, transaction characteristics influence price and
dealer profits. Whether the vehicle was leased, or financed, or if the vehicle was bought at
the end of the month, or year, and whether there was a trade-in vehicle associated with the
transaction can have important effects. If gender and age categories exhibit systematic variation over transaction characteristics, omitting them would bias our results. Such variables,
or suitable proxies, are included as controls.
We adapt the reduced form from Busse et al. (2013) to estimate a simple linear model
of dealer profits from a particular vehicle transaction. We denote a model-trim combination
by m, model-year by y, province by p, date by t. Then Dealer d’s profits from a transaction
with customer i is:
9
Segment
Compact (37%)
Full (0%)
Luxury (4%)
Midsize (13%)
Pickup (13%)
SUV (20%)
Sports (1%)
Van (9%)
Total (100%)
Male Cust.
Female Cust.
Cust.
Price
Dealer
Marg
Cust.
Price
Dealer
Marg
$20,183
$35,342
$48,590
$28,844
$39,446
$36,263
$45,141
$28,446
$31,075
$1,327
$1,217
$2,740
$1,650
$2,202
$1,957
$3,335
$1,337
$1,784
$19,720
$35,108
$45,773
$28,247
$38,678
$34,496
$37,627
$28,377
$26,589
$1,302
$1,164
$2,614
$1,643
$2,240
$1,922
$2,542
$1,365
$1,590
Source: Authors’ Calculations
Table 5: Mean Customer Price and Dealer Margin by Gender.
id
πmypt
= β0 + β1 demogi + β2 transcharidt + λd + µt + Θmpy + idt
(1)
Specifically, the dealer’s profit depends on a) the demographics of the customer, captured
by age and gender; b) characteristics of the transaction — which includes timing-related
measures such as whether the vehicle sold on a weekend, the end of a month or the end
of a year — and characteristics of the purchase such as whether the vehicle was leased,
financed or bought with cash; c) dealer fixed effects, denoted λd ; d) year-month fixed effects,
denoted µt ; e) characteristics of the vehicle, denoted Θmpy Estimating this equation yields
the average difference in dealer margin within model-province-year-trim across our gender
and age categories.
4
Regression Results
We now present estimation results for variants of Equation 1. Table 6 presents results from a
set of regressions where the dependent variable is the dealer’s margin on each transaction.15
We do not transform the dealer margin by taking a log, as this would require dropping
the 8% of observations where margins are negative.16 We estimate the relationship between the dealer’s margin and the age and gender of consumers. The omitted age group
is consumers below 25. Column 1 includes fixed-effects for the province of the transaction,
Column 2 uses Model*model-year*province fixed-effects, and Column 3 uses Model*modelyear*trim*province fixed-effects. All regressions also include year*month fixed-effects to
control for general trends in vehicle costs and prices.
15
Dealer margin is the difference between transacted price and vehicle cost.
Dealers make negative margins for a number of reasons, including responding to manufacturer incentives,
getting rid of cars with low demand, taking losses on the new car in order to make profits on the trade-in, or
to generate future servicing incomes. Later, we show that our results are robust to using only transactions
with positive dealer margins and expressing the dependent variable in logs.
16
10
Figure 2: Customer Age Distribution for Popular Models
Civic
Corolla
Escape
Yaris
4000
3000
Number Sold
2000
1000
0
4000
3000
2000
1000
0
25
50
75
25
50
75
Customer Age
The results in the first column show that dealers, on average, make $107 less from female
consumers. as well as lower margins on younger consumers. 45 to 65 year old consumers
generate approximately $280 more in margins than the omitted group of consumers under
25, and about $100 more than those between 25 and 30.
Note, however, that since Column 1 does not control for the make, model or trim of the
vehicle, the results for different age and gender groups may be driven by the types of cars
they purchase. For example, if female consumers tend to buy cheaper models or trims, and
if dealers make lower margins on cheaper cars, then the results may just reflect these choices
rather than differences in negotiation. In Table 9 in Appendix A.1 we show that this is
indeed the case.17
Therefore, in column 2 of Table 6, we estimate the same relationship as before, but using
model*model-year*province fixed-effects. We now find that dealers earn only slightly higher
margins from female consumers, but considerably higher margins from older consumers—as
much as $100–130 from consumers over 65. In column 3, we estimate this relationship within
a vehicle’s model-trim. We find that the differences persist, and are in fact somewhat larger.
Consumers above 65 generate about $105–144 higher margins for dealers than consumers
under the age of 35.
Within model and model-trim there is less room for vehicle cost to vary. Within models,
customers can choose more expensive trims, or additional options (the only possibility within
trims). As we show in Table 9 in Appendix A.1, age and gender related differences in vehicle
17
That Table shows that the youngest consumers buy the cheapest cars. Consumers aged 25-30 buy 10%
more expensive vehicles on average, and the premium is about 18–20% for the other age groups. Female
customers, on average purchase vehicles that are 12% cheaper than males. Thus, the observed differences in
Column 1 of Table 6 are related to the differences in the choice of vehicles across gender and age categories.
11
Table 6: Regression of Dealer Margin
Prov
Model-Prov
Model-Trim-Prov
a
a
Female
-107.6
(3.9)
13.0
(3.6)
21.5a
(3.6)
a
Age 25-30
107.5
(10.0)
6.5
(9.1)
8.4
(9.0)
a
Age 30-35
178.0
(9.8)
6.7
(9.0)
8.5
(9.0)
a
a
a
Age 35-40
224.6
(9.6)
23.2
(8.9)
26.5
(8.9)
Age 40-45
254.7a
(9.3)
45.1a
(8.6)
48.1a
(8.6)
a
a
a
Age 45-50
269.5
(9.2)
68.6
(8.5)
71.8
(8.5)
a
a
a
Age 50-55
281.2
(9.4)
84.8
(8.6)
87.4
(8.6)
Age 55-60
282.7a
(9.8)
84.5a
(9.0)
88.9a
(9.0)
a
a
a
Age 60-65
257.4
(10.6) 79.5
(9.7)
83.9
(9.7)
Age 65-70
232.3a (11.7) 98.6a
(10.8) 105.2a
(10.8)
Age > 70
194.0a (10.8) 130.1a (10.1) 144.4a
(10.1)
a
a
a
Financed Indicator -44.0
(6.1) 171.5
(5.8) 180.4
(5.8)
Leased Indicator
36.3a
(6.0) 185.5a
(5.7) 198.0a
(5.7)
a
a
a
Sat or Sun FE
20.4
(6.1)
22.3
(5.5)
23.1
(5.5)
End of Month FE
-40.5a
(4.4)
-60.0a
(4.0)
-61.0a
(4.0)
End of Year FE
-88.1a (17.9) -93.2a (16.2) -93.6a
(16.2)
a
a
a
Constant
1607.8 (31.0) 2384.5 (26.7) 2360.3
(26.9)
R2
0.037
0.239
0.264
c
p < 0.1, b p < 0.05, a p < 0.01. Fixed-effects specified as Col. 1: Province; Col. 2:
Model-Year-Province; Col. 3: Model-Year-Trim-Province. All regressions include yearmonth FEs. N=510866.
cost are very small within a given model and trim.18 Thus, once the choice of model is
controlled for, as it is in Columns 2 and 3 of Table 6, differences in dealer margins must
emerge from the negotiation process rather than from different vehicle choices by consumers,
as we discuss in more detail below.
In Table 7 we use the joint distribution of age and gender categories. The omitted
category is male customers under the age of 25. Dealers make their lowest margins on
young, female consumers. This is most obvious in column 1, where women under 25 generate
almost $300 lower margins than men of the same age, and about $450 lower margins than
men aged 50–60. But this is partly driven by the choice of cars that these types of consumers
purchase.19
When we control for the model in Column 2, women under 25 pay about $140 less than
the omitted group, which is men of the same age. The group generating the greatest profit
for dealers is women above the age of 70, who pay an average of $85 above the omitted
group. In column 3 we control for model-trim as well and see the same pattern of results.
18
For example, column 2 of Table 9, which is estimated within models, shows that female customers do
not, on average, purchase more expensive trims, or greater options than male customers.
19
From the corresponding vehicle cost regression (column 1 of Table 10, Appendix A.1), we see that in
every age category, female consumers buy cheaper vehicles on average than male consumers. The malefemale new car premium is roughly 13–16% in most age groups. Across all age and gender categories, young
women buy the cheapest cars, while 35 to 65 year old men buy the most expensive cars.
12
Table 7: Regression of Dealer Margin
Prov
Model-Prov
Model-Trim-Prov
a
a
Age < 25 Female
-291.6 (15.2) -138.4 (13.8) -115.3a
(13.7)
Age 25-30 Male
58.2a
(14.2) -34.7a (12.8) -26.7b
(12.8)
a
a
a
Age 25-30 Female
-148.0 (14.4) -101.4 (13.1) -82.1
(13.1)
a
a
a
Age 30-35 Male
92.0
(13.6) -58.2
(12.4) -49.3
(12.4)
Age 30-35 Female
-40.1a (14.5) -79.7a (13.2) -61.8a
(13.2)
a
a
a
Age 35-40 Male
127.4
(13.3) -54.7
(12.2) -44.1
(12.2)
Age 35-40 Female
22.1
(14.4) -45.9a (13.1) -26.7b
(13.1)
a
a
a
Age 40-45 Male
146.9
(13.1) -43.8
(12.0) -31.8
(12.0)
a
Age 40-45 Female
66.5
(13.8)
-8.9
(12.6)
7.8
(12.6)
Age 45-50 Male
159.8a (13.0) -25.2b (11.8)
-13.5
(11.9)
a
c
a
Age 45-50 Female
83.6
(13.6)
21.1
(12.4) 38.5
(12.4)
Age 50-55 Male
168.0a (13.1)
-6.2
(12.0)
4.3
(12.0)
a
a
a
Age 50-55 Female
100.6
(13.9) 34.0
(12.7) 51.5
(12.7)
a
Age 55-60 Male
166.7
(13.5)
-15.6
(12.4)
-2.3
(12.4)
Age 55-60 Female
108.1a (14.6) 48.0a
(13.3) 65.9a
(13.3)
Age 60-65 Male
154.2a (14.3)
-13.6
(13.1)
-3.3
(13.1)
a
b
a
Age 60-65 Female
65.0
(16.2)
34.9
(14.7) 57.2
(14.7)
Age 65-70 Male
128.2a (15.6)
13.3
(14.2)
25.9c
(14.2)
b
b
a
Age 65-70 Female
42.4
(18.4)
42.0
(16.7) 66.2
(16.7)
Age > 70 Male
79.2a
(14.3) 40.1a
(13.2) 63.0a
(13.3)
c
a
a
Age > 70 Female
29.5
(17.3) 85.0
(15.8) 112.2
(15.8)
a
a
a
Financed Indicator -43.6
(6.1) 172.0
(5.8) 180.9
(5.8)
Leased Indicator
37.2a
(6.0) 186.3a
(5.7) 198.7a
(5.7)
Sat or Sun FE
20.2a
(6.1)
22.3a
(5.5)
23.1a
(5.5)
a
a
End of Month FE
-40.4
(4.4)
-59.8
(4.0)
-60.9a
(4.0)
a
a
a
End of Year FE
-87.9
(17.9) -93.1
(16.2) -93.6
(16.2)
a
a
a
Constant
1703.5 (31.9) 2463.4 (27.5) 2432.0
(27.7)
R2
0.038
0.240
0.264
c
p < 0.1, b p < 0.05, a p < 0.01. Fixed-effects specified as Col. 1: Province; Col. 2:
Model-Year-Province; Col. 3: Model-Year-Trim-Province. All regressions include yearmonth FEs. N=510866.
13
Figure 3: Dealer Margin across demographic groups, relative to Men under 25
150
Dealer Margin
100
50
Gender
female
0
male
−50
−100
−150
<25
25−30 30−35 35−40 40−45 45−50 50−55 55−60 60−65 65−70
70+
Age Group
The lowest profits for dealers come from women under 25, followed by those between 25 and
30. The highest profits are generated on women above 70, followed by those in the 60–65
age group.
In general, the results show that dealer margins rise with the age of consumers, as we see
that older men also pay more than average. However, the striking result is that, while young
women do better than men of the same age, older women do considerably worse. Thus, not
only does the premium paid by consumers rise with age, but it does so much more steeply
for women than for men. We illustrate this in Figure 3, by plotting the coefficient estimates
from Column 3 of Table 7. The shaded regions around each plot are the 95% confidence
intervals.20
Using the coefficients from column 3, dealer margins are $115 lower for young (under 25)
female customers, but over $110 higher for older (70-75) female customers, relative to the
omitted group of men under 25. The average dealer margin earned across all vehicles in our
data is $1705. Thus, old women generate a margin that is 13% higher than younger women
buying the same vehicle. In the context of the purchase of a car that costs over $20,000,
these amounts might not seem much. However, note that these are average amounts and
that some consumers can — and do — save more through a combination of careful research,
obtaining multiple quotes, and adopting different negotiation strategies.
Turning to the other factors that may influence dealer margins, we see from the third
column in both Tables 6 and 7 that dealers make $180–200 higher margins on cars that
20
The figure shows that the coefficients for men and women of the same age are usually significantly
different from each other, except in the 30–40 age groups where the lines cross over. While the confidence
intervals overlap slightly among consumers aged 65–70, the coefficient estimates for this age group continue
to be significantly different from each other. Indeed, with the exception of consumers in the 30–40 age range
where the lines cross, the gender gap (difference in coefficients) is significant at the 1% level for all age
groups. Later, we address the issue of how to cluster standard errors.
14
are leased or financed, relative to outright cash purchases. This accords with common
observations about the new car market. Finally, dealers clearly sacrifice margins at the end
of the month and year. On average, their margins are $61 and $94 lower at these times, again
conforming to casual observations, and confirming that dealers respond to manufacturer
incentives at the end of calendar months and years. Note that these lower margins may
either be due to dealers being willing to sacrifice profits to to meet sales quotas, or by more
price-conscious consumers choosing to purchase cars at these times, knowing the incentives
dealers face.
4.1
Robustness
In this section we show that our results are not driven by other factors, such as omitted
variables, outlier observations, or selection. We present the main coefficients of interest
graphically in Figures 4 and 5. In each figure, the solid blue line represents dealer margins
on female customers relative to men under 25, while the dashed red line corresponds to
margins on male customers. Note that these results are obtained from regressions using
the same specification as Column 3 of Table 7, which include fixed effects for model-yearprovince-trim and for year-months.21 Regression results, and the full set of standard errors
for these exercises, are presented in Tables 11, 12, 13 and 14 in the Appendix.
We start by examining particular segments — Compact cars, SUVs and Minivans in
Panels A and B of Figure 4. These segments are disproportionately associated with certain
age groups, as we saw in Section 2; younger consumers are more likely to purchase small
cars, while middle-aged consumers, especially those that are married, are more likely to
purchase family cars such as SUVs and minivans. We then show, in Panel C, that our
results are robust to examining popular car models alone; this is to verify that are results
are not driven by unusual consumer or dealer behavior in vehicles with low sales. We define
high-selling car models as those with at least 5000 units in sales during our sample period.
This corresponds to 26 model-years, out of our observed set of more than 500 models, with
almost 50% of total vehicle sales.
Next, we restrict the sample, in Panel D, to those where dealers make positive margins on
new car sales. While we have explained above that there may be rational reasons for dealers
to make losses on certain transactions, one may be concerned that these sales are specific
to certain types of cars or to unobserved characteristics of the transaction which may be
correlated with consumer demographics. We then examine domestic (North American) and
foreign manufacturers separately in panels E and F. Anecdotal reports suggest that dealers
of foreign cars are less likely to negotiate on final prices; if so, this may affect our results if
demographic groups differ in their propensity to purchase foreign cars.
Next, we consider the possibility that new vehicle transaction prices may be influenced
by the amounts negotiated on consumers’ trade-in vehicles. While in principle the two
transactions should be treated separately, in practice dealers may allow overpayment on some
trade-ins in order to make greater margins on the new car, or vice versa.22 We therefore
21
We do not present confidence intervals in these figures to avoid crowding. Later, we discuss the appropriate level of clustering standard errors, and present tests of significance.
22
See Zhu et al. (2008) and Busse and Silva-Risso (2010) for evidence regarding this possibility.
15
Figure 4: Dealer Margins for subsamples, by Gender and Age
A: Compacts
B: SUVs+Vans
C: High−selling
100
Dealer Margin
0
−100
Gender
−200
D: Positive Margins
E: Domestic
female
F: Foreign
male
100
0
−100
−200
25−30 40−45 55−60
70+ 25−30 40−45 55−60
70+ 25−30 40−45 55−60
70+
Age Group
restrict attention, in Panel A of Figure 5, to the subset of transactions where consumers did
not trade-in an older vehicle.
We now control for the possibility that leased or financed vehicles may affect the results,
even though we have included controls for such transactions in our regressions. Harless and
Hoffer (2002) did not consider leased cars in their study of dealer transactions, arguing that
there is generally little room to negotiate prices in such cases. Additionally, transactions
that involve financing through the dealer may be problematic if, for example, the dealer is
willing to accept a lower price on the car in return for higher interest payments. Therefore,
in Panels B and C of Figure 5 we drop observations which involve leased cars and vehicle
financing, respectively.
The exercises so far have involved different cuts of the data. We now perform three
additional checks where we add control variables that may affect the results, using the full
sample that was employed in Table 7. In Panel D of Figure 5 we include the cost of the vehicle
as a regressor. Although the model*year*trim fixed-effects mostly control for the value of
the car, they do not include the purchase of aftermarket options and service contracts, which
are likely to be correlated with dealer margins.
Next, we control for systematic pricing behavior by certain dealers. Suppose, for example,
that dealers located in smaller cities or suburban locations charge lower prices for exactly
the same car as a city dealer who faces higher costs. If the demographic distribution of
consumers in these two locations is also different—for example, if older consumers are more
likely to live near high-cost dealers—then this selection of consumer types may drive our
16
Figure 5: Dealer Margins for subsamples, by Gender and Age
A: No Trade−ins
B: No Leases
C: No Financing
100
50
0
Dealer Margin
−50
−100
Gender
D: Vehicle Cost
E: Dealer Controls
female
F: Days to Turn
male
100
50
0
−50
−100
25−30 40−45 55−60
70+ 25−30 40−45 55−60
70+ 25−30 40−45 55−60
70+
Age Group
results. We control for this possibility in Panel E by including dealer fixed-effects, which
allow us to look within dealers. Finally, in Panel F, we include the number of days that
the vehicle has been on the dealer’s lot—popular cars typically turnover very quickly and so
dealers may be willing to reduce margins on cars that are not in high demand.
4.2
Tests of Significance and Clustering
We now discuss the significance of our results in the context of appropriate clustering of
standard errors. The standard errors presented in Tables 6 and 7 were not adjusted for
clustering. This may understate the standard errors if there are common shocks across
observations drawn from certain groups. We consider three types of groups that may give
rise to such common shocks, and therefore three ways by which to cluster standard errors.
The first is by model-year: if demand for certain model-years is correlated across consumers then this may be reflected in the margins obtained by dealers on such transactions.
The second is by year-month: our data include the period following the 2008 financial crisis,
when the automobile industry in North America was contracting. There is a strong likelihood dealers made systematically lower margins during this period. The third is by dealer:
vehicles sold by the same dealer are subject to the same supply side behavior, and thus
possibly correlated final prices.
In Table 8 we list the results from hypothesis tests for our main findings using four
different types of clustering: none, and by dealer, model-year, and year-month. The table
17
Table 8: Significance Tests for the Gender Gap in certain age groups
Age Group
Premium
P-value according to clustering:
Under
25 to
65 to
Over
25
30
70
70
(Female-Male) None Dealer
-115.3 0.000 0.000
-55.5 0.000 0.000
40.3 0.014 0.018
49.2 0.001 0.000
Model-Year
0.000
0.000
0.021
0.000
Year-Month
0.000
0.000
0.016
0.000
shows that the gender gap is negative and highly significant among the two youngest age
categories, and positive and highly significant among the two oldest categories. The results
are significant at the 99% level in three cases, and at the 95% level in the case of the 65-70
age group.23 Clustering standard errors has little or no effect on the p-values.
We constructed similar tables for the robustness regressions used in Section 4.1; see
Appendix Table 15 for a few examples. In all cases, the results that older women pay more
than younger women, and that younger women pay less than younger men, continue to hold
and are highly significant. The gender gap among older consumers is not always statistically
significant; for example in the case of Compact cars as had been mentioned above. This
is not entirely surprising since a number of robustness exercises use a considerably smaller
sample, which leads to large confidence intervals. In any case, the method of clustering
appears to have little effect on the standard errors.
4.3
Summary
We emphasize three findings emerging from our results—see the original Figure 3, and all
twelve robustness panels from Figures 4 and 5. a) There is a clear age premium among
women: women in the oldest age category generate over $200 higher margins for dealers
than those in the youngest age category, and there is a steady, almost monotonic, increase
in margins with age. b) The age premium rises more steeply for women than for men—in all
twelve panels, the solid line corresponding to the female premium starts below the dashed line
corresponding to the male premium, and ends up above it. Younger women outperform men
of the same age by about $100, while older women do worse than their male contemporaries
by about $50 on average, subject to exceptions discussed below. c) There is no obvious age
premium among men, although there are certain patterns that appear common across most
panels. The very youngest male consumers — those under 25 — pay a premium relative to
men in the next two age categories. Meanwhile, men above the age of 65 also appear to pay
a premium relative to younger men, similar to the case of older women. In general, though,
the margin–age relationship for men appears much flatter than for women, which fits with
our view of treating male customers as a control group.
These three results are less robust in some subsamples. For example, the gender gap
among older consumers is quite small, and not statistically significant, when we examine
compact cars and foreign cars in Panels A and F, respectively, of Figure 4 (see the Appendix
23
Note that these are all two-sided tests.
18
tables for detailed results and standard errors). Generally, though, the finding of gender
differences among older and younger consumers is robust.
These results are interesting and also surprising. We control for the model, province and
trim of each vehicle (using various fixed effects). We control for the timing of purchase,
and in robustness exercises control for location of purchase, and the dealer’s cost for the
vehicle transacted. We examine differences in dealer margins within as similar a product
as possible—almost the same car being sold at the same time and place—across consumers
of different demographic characteristics. Therefore, any remaining differences in margins
derive solely from demographics. Do dealers and customers negotiate differently as consumer
demographics vary?
5
Reasons for disparities across Gender and Age
Our results—Figures 3, 4, and 5—demonstrate a clear market disparity across gender and
age. Note that we see no evidence of discrimination.24 For example, there is no evidence
that dealers treat women differently in the aggregate: while older women do perform worse
in negotiations, younger women perform very well.25
Our data is a snapshot of the new car market (all transactions take place during 2004–
2009). We do not follow cohorts over time. Thus, one possibility is that our findings illustrate
an ongoing effect, where the same consumers fare progressively worse in the new car market
as they age. Another possibility is that they illustrate a cohort effect—the current cohort
of young women possess a superior ability to negotiate with car dealers than men of their
generation. Either way, we need to explain why the relationship between age and dealer
margin is steeper for female rather than male customers.
Consider first the ‘ongoing’ effect. Prior research (Morton et al. (2011)) demonstrates the importance of search and negotiation costs in buying new cars. Morton et al.
(2001) illustrate how access to information from online car referral services reduce the price
paid by consumers. In Table 20 we present the results of Statistics Canada’s Internet use
survey (held every two years) for the period 2005-2012. Most individuals younger than 34
years of age (over 88% for all columns) use the Internet for personal use, as opposed to most
individuals aged 65 and older (below 47.5% and as low as 23.8%). Prior research on US
residents also indicates that the likelihood of Internet use declines with age; see Goldfarb
and Prince (2008). If older consumers are less likely to access the Internet to research their
vehicles, they will have higher search and negotiating costs. If older consumers are always
less likely to adopt new technologies, they will always do worse in market outcomes as they
24
According to Heckman (1998), “[d]iscrimination is said to arise if an otherwise identical person is treated
differently by virtue of that person’s race or gender, and race and gender by themselves have no direct effect
on productivity.”
25
Even if discrimination creates (at least a portion of) the observed disparity across certain groups, it
is unclear whether this would be ‘statistical’ or ‘taste’ based. If older female customers on average have
higher costs of negotiation than males of the same age (see Babcock and Laschever (2003)), a dealer’s belief
in the distribution of negotiating skills across age and gender would augment an already unequal market.
Perceived differences in search costs across age and gender—through Internet access, or a propensity to
research vehicles—would have the same effect (as explained in Morton et al. (2003)).
19
age.26
Lambert-Pandraud et al. (2005) find evidence that older customers repurchase brands
more frequently on purchasing their vehicles. They also consider fewer brands, fewer dealers, and fewer models than younger customers. In Lambert-Pandraud and Laurent (2010)
the authors find the behaviour repeating itself in the french perfume market. Attachment
to brands can allow dealers to extract higher profits from older customers. Both these studies offer several explanations for age related brand attachment including: change aversion,
cognitive decline, and even nostalgia.
The possibilities above are useful in explaining a relationship, across both genders, between age and dealer margin. However, they cannot explain why the female relationship is
steeper.
The literature on ‘sensation seeking,’27 see Wong and Carducci (1991) explores gender
differences. Males typically outscore females in sensation seeking. They also find that
sensation seeking declines as people age. It is plausible that young, impetuous men, seeking
‘sensation’ from owning their vehicles, under-invest in research or in obtaining multiple
quotes, which would affect their negotiation strategies. However, as these men mature, they
may become less impulsive in their purchasing behavior. This would explain why men under
30 in our data generate higher dealer margins for the same vehicles than males in the next
two age categories. Nevertheless, this literature also cannot explain why older women do
worse than older men.
Now consider the ‘cohort’ effect. Our understanding of ongoing effects cannot explain why the female relationship between age and dealer margin is steeper than the male
one. So we consider a ‘cohort’ effect instead.
The labor market experience and educational attainment of women for those under 30
today is starkly different from those in their 60s and above. Over the past few decades,
women in North America have considerably narrowed their gap in education with men, and
in recent years have even surpassed men (see Goldin et al. (2006) for evidence for evidence
from the US). In Canada, young women are much more likely to have completed tertiary
education than older women.28 Younger women are also more likely to have completed
tertiary education than men (with their population shares being 10-16% greater for the
period 2000-2011). At the same time, women’s labor force participation has grown rapidly.
In Table 17 we present a comparison of female employment rates across cohorts. In the
year 2006, women aged 25-29 were much more likely to be employed than those aged 55-59
when that cohort was the same age (we are comparing employment outcomes of those born
between year 1976-80, with those born between 1946-50). By contrast, in 2006, men aged
25-29 were less likely to be employed than those aged 55-59 when the latter cohort was the
same age (see male employment rates in Table 18).
The economic effects of these societal changes can potentially explain our findings. Most
26
Alternatively, older consumers might value their time higher than younger consumers—potentially due
to higher wages. However this is inconsistent for the retired who generally earn lower wages and thus have
lower time costs.
27
Sensation seeking is defined as “the need for varied, novel, and complex sensation and experiences and
the willingness to take physical and social risks for the sake of such experiences” Zuckerman (1979) p. 10.
28
Specifically, 62% of women aged 25-34 have university degrees as compared to 38% of women aged 55–64.
See 19 and its accompanying discussion.
20
prior research on these economic consequences has focused on labor market outcomes—for
example, demonstrating reductions in the gender pay gap (see Goldin (2006)). There is
considerably less research on its impact on other economic outcomes, particularly on price
negotiations. In contrast, men have not undergone such transformative societal changes. If
we consider male consumers as a control group, then our results provide evidence that young
women today are fundamentally different from the older women in our sample.
Our results are consistent with the notion that the better educational attainment and
labor force participation of young women, relative to men of the same age, allows them to
perform better in price negotiations. Women above the age of 65 are much less likely to have
completed high-school or college, or to have been employed full-time when they were younger.
These younger women of today are better negotiators than their male counterparts. If this is
due to their superior educational attainment, and labor market outcomes it is unlikely that
they will do worse than their male counterparts as they age. The difference in relationships
observed is therefore likely to be particular to the cohorts currently observed, rather than
reflecting an ongoing effect.
5.1
Cross-Province Differences
In this section we explore the following hypothesis: if cohort effects explain our observed
gender gap, these effects should vary across provinces with varying gender differences in
employment and education.
There are, however two obstacles to systematically studying cross-province differences.
First, Canada’s provinces have shown similar improvements in the gender gap over time,
implying that there is little cross-province variation to exploit in terms of education and employment. In this regard, data from the United States would have been more desirable, had
it been available, due to much greater variation among the states and regions in that country.
Canadian provinces are more homogeneous in terms of socio-economic characteristics and
therefore regional differences are less apparent.29
The second obstacle is that many Canadian provinces are small, and the number of
new cars sold each year is too low in most cases to permit consistent estimates of gender
differences in negotiated prices. We are thus able to estimate coefficients for Canada’s two
largest provinces of Ontario and Quebec, but not for the others.
In Figure 6 we present the results of a cross-province analysis using Ontario and Quebec. The two top figures show the estimated coefficients from estimating the dealer margin
regressions separately for transactions in each province; we use the same specification as
in Column 3 of Table 7, which includes fixed effects for model-year-province-trim and for
year-months.30 These figures confirm the general result for the country as a whole, with a
clear age premium for women, and with older and younger women paying the highest and
lowest prices, respectively, among all consumers. However, comparing the coefficients across
provinces suggests that women in Quebec perform better, relative to men, than in Ontario.
The solid blue line crosses the dashed red line at a lower age group in Ontario than in Quebec,
29
To be clear, there has been significant variation in these characteristics along other dimensions such as
by race and across the rural-urban divide. However, our transactions data are at the provincial level, and
average changes among provinces has been quite similar.
30
See Table 16 in the Appendix for detailed results.
21
Figure 6: Negotiated prices, education and employment: Ontario and Quebec
Ontario
Quebec
Dealer Margin
200
100
Gender
female
male
0
−100
25−30
40−45
55−60
70+
25−30
40−45
55−60
70+
12
University Gap (Men−Women)
Employment Gap (Men−Women)
Age Group
8
4
0
−4
25−30
35−40
45−50
55−60
65+
0.10
0.05
province
Ontario
Quebec
0.00
−0.05
25−44 45−54 55−64 65+
Age
Age
suggesting that women as old as in their 60s outperform men in Quebec, while this is only
true for women up to their 30s in Ontario. Additionally, the gender gap is generally much
more favorable to women in Quebec than in Ontario: it is larger among young consumers,
and smaller among older consumers. For example, women under 25 outperform their male
counterparts by about $120 in Quebec, but only about $80 in Ontario.
The two lower figures show the gender gap in employment and education for these two
provinces.31 In each case, we construct the difference between the relevant measure for men
and for women in each province. These figures show that women in Quebec have more
favorable socioeconomic outcomes, relative to men, than in Ontario. The youngest cohort of
women in each province are both more likely to have attained a university degree and more
likely to be employed than men, but the gap is bigger in Quebec. In the 30-50 age groups,
men generally outperform women but by less in Quebec than in Ontario. The difference in
the gender gap across provinces can be sizable: women in their 30s are 5 to 6 percentage
points more likely to be employed, relative to men, in Quebec than in Ontario. These figures
are similar if we use alternative measures of employment and education; for example, labor
31
The data were obtained from Statistics Canada. Although employment and labor force participation
data are available for 5-year age categories at the province level, we were unable to obtain data on educational
attainment at a more disaggregated level than shown.
22
force participation or the fraction of the population with less than a high-school diploma.
These results are obtained using just two provinces, and we do not claim that this evidence is conclusive. Nevertheless, these comparisons strongly suggest that improvements in
women’s education and labor force participation have contributed to improvements in their
ability to negotiate prices.
6
Conclusions
In this paper we examine the gender gap in price negotiations in the new car market. Our
results provide clear evidence of a reversal in the gender gap. In particular, while women
aged 65 and older fare worse in negotiations than men of the same age, the gap narrows
and then changes direction among younger consumers. Women under the age of 30 perform
significantly better than their male peers.
These findings are concurrent with the reversal of the gender gap in education, and the
narrowing of the gender gap in employment and wages. We consider various explanations
for our findings and conclude that the most likely explanation is that the rapid increase
in women’s education, as well as the improvement in their earnings and work experience
relative to men, has given women better information and more confidence while conducting
price negotiations. We find supporting evidence for this by examining changes in education
and employment in Canada’s two largest provinces, and comparing these to differences in
negotiated prices.
The main limitation of our findings is that they are based on a cross-section of transactions, albeit one that employs a very large sample of consumers drawn from the entire
range of the age distribution. Confirmation of our results requires a panel study that tracks
cohorts over a much longer period. Since such data do not currently exist in the automobile
industry, our findings should be regarded as suggestive until they can be confirmed in the
future.
Our results imply that the improvements women have seen from more advanced education
and greater work opportunities have not been restricted to the labor market, but potentially
extend to other settings too. This has important implications for the housing market, in
particular, which also features price negotiations and comprises the single biggest financial
transaction of many consumers’ lives.
References
Ayres, I. and P. Siegelman (1995, June). Race and gender discrimination in bargaining for
a new car. American Economic Review 85 (3), 304–21.
Babcock, L. R. and S. Laschever (2003). Women Don’t Ask: Negotiation and the Gender
Divide. Princeton University Press.
Bertrand, M., C. Goldin, and L. F. Katz (2010). Dynamics of the gender gap for young
professionals in the financial and corporate sectors. American Economic Journal: Applied
Economics, 228–255.
23
Bertrand, M. and K. F. Hallock (2001, October). The gender gap in top corporate jobs.
Industrial and Labor Relations Review 55 (1), 3–21.
Blau, F. D. and L. M. Kahn (2007). The gender pay gap have women gone as far as they
can? The Academy of Management Perspectives 21 (1), 7–23.
Busse, M. R., C. R. Knittel, and F. Zettelmeyer (2013). Are consumers myopic? evidence
from new and used car purchases. The American Economic Review 103 (1), 220–256.
Busse, M. R. and J. M. Silva-Risso (2010). ” one discriminatory rent” or” double jeopardy”:
Multicomponent negotiation for new car purchases. The American Economic Review ,
470–474.
Goldberg, P. K. (1996, June). Dealer price discrimination in new car purchases: Evidence
from the consumer expenditure survey. Journal of Political Economy 104 (3), 622–54.
Goldfarb, A. and J. Prince (2008). Internet adoption and usage patterns are different:
Implications for the digital divide. Information Economics and Policy 20 (1), 2–15.
Goldin, C. (2006). The quiet revolution that transformed women’s employment, education,
and family. American Economic Review 96 (2), 1–21.
Goldin, C., L. F. Katz, and I. Kuziemko (2006). The homecoming of american college women:
The reversal of the college gender gap. Journal of Economic Perspectives 20 (4), 133–156.
Harless, D. W. and G. E. Hoffer (2002, March). Do women pay more for new vehicles?
evidence from transaction price data. American Economic Review 92 (1), 270–279.
Heckman, J. J. (1998). Detecting discrimination.
tives 12 (2), pp. 101–116.
The Journal of Economic Perspec-
Lambert-Pandraud, R. and G. Laurent (2010). Why do older consumers buy older brands?
the role of attachment and declining innovativeness. Journal of Marketing 74 (5), 104–121.
Lambert-Pandraud, R., G. Laurent, and E. Lapersonne (2005). Repeat purchasing of new
automobiles by older consumers: empirical evidence and interpretations. Journal of Marketing 69 (2), 97–113.
Langer, A. (2011). Demographic preferences and price discrimination in new vehicle sales.
manuscript, University of Michigan.
Morton, F. S., J. Silva-Risso, and F. Zettelmeyer (2011). What matters in a price negotiation:
Evidence from the us auto retailing industry. Quantitative Marketing and Economics 9 (4),
365–402.
Morton, F. S., F. Zettelmeyer, and J. Silva-Risso (2001). Internet car retailing. The Journal
of Industrial Economics 49 (4), 501–519.
24
Morton, F. S., F. Zettelmeyer, and J. Silva-Risso (2003). Consumer information and discrimination: Does the internet affect the pricing of new cars to women and minorities?
Quantitative Marketing and Economics 1 (1), 65–92.
O’Neill, J. (2003). The gender gap in wages, circa 2000. The American Economic Review 93 (2), 309–314.
Stuhlmacher, A. F. and A. E. Walters (1999). Gender differences in negotiation outcome: A
meta-analysis. Personnel Psychology 52 (3), 653–677.
Turcotte, M. (2011). Women and education. Women in Canada: A Gender Based Statistical
Report.
Wong, A. and B. J. Carducci (1991). Sensation seeking and financial risk taking in everyday
money matters. Journal of Business and Psychology 5 (4), 525–530.
Zhu, R., X. Chen, and S. Dasgupta (2008). Can trade-ins hurt you? exploring the effect
of a trade-in on consumers’ willingness to pay for a new product. Journal of Marketing
Research 45 (2), 159–170.
Zuckerman, M. (1979). Sensation seeking: Beyond the optimal level of arousal (1st ed.).
Hillsdale, NJ: Lawrence Erlbaum Associates, Publishers.
25
A
Appendix (Not For Publication)
A.1
Vehicle Cost Regressions
We first examine how the age and gender of consumers relates to vehicle cost. In Table 9 we present the results from a regression of the log of the vehicle cost, 32 on demographic characteristics. Column 1 includes fixed-effects for the province of the transaction.
Column 2 uses Model*model-year*province fixed-effects, and Column 3 uses Model*modelyear*trim*province fixed-effects. All regressions also include year-month fixed-effects to control for general trends in vehicle costs and prices. In the first column we estimate the relaTable 9: Regression of Log(Vehicle Cost)
Prov
Model-Prov
Model-Trim-Prov
a
Female
-0.123 (0.001) -0.000 (0.000) 0.003a (0.000)
Age 25-30
0.091a (0.002) 0.003a (0.001) 0.003a (0.001)
Age 30-35
0.162a (0.002) 0.005a (0.001) 0.006a (0.001)
Age 35-40
0.201a (0.002) 0.007a (0.001) 0.008a (0.001)
Age 40-45
0.202a (0.002) 0.007a (0.001) 0.008a (0.001)
Age 45-50
0.188a (0.002) 0.008a (0.001) 0.008a (0.001)
Age 50-55
0.180a (0.002) 0.009a (0.001) 0.009a (0.001)
Age 55-60
0.184a (0.002) 0.010a (0.001) 0.010a (0.001)
Age 60-65
0.183a (0.003) 0.009a (0.001) 0.011a (0.001)
Age 65-70
0.160a (0.003) 0.005a (0.001) 0.010a (0.001)
Age > 70
0.117a (0.003) -0.000 (0.001) 0.008a (0.001)
Financed Indicator -0.033a (0.001) 0.019a (0.001) 0.022a (0.000)
Leased Indicator
0.031a (0.001) 0.027a (0.001) 0.029a (0.000)
Sat or Sun FE
-0.005a (0.001) -0.001a (0.001) -0.001b (0.000)
End of Month FE
0.015a (0.001) -0.001a (0.000) -0.001a (0.000)
End of Year FE
0.010b (0.004) -0.001 (0.002) -0.001 (0.001)
Constant
9.987a (0.007) 10.223a (0.003) 10.199a (0.002)
R2
0.171
0.896
0.935
c
p < 0.1, b p < 0.05, a p < 0.01. Fixed-effects specified as Col. 1: Province; Col. 2:
Model-Year-Province; Col. 3: Model-Year-Trim-Province. All regressions include yearmonth FEs. N=510866.
tionship between age, gender, and the cost of a new vehicle purchased while allowing make,
model, and trim to vary. We always omit consumers below 25 (for all columns). The results
in column 1 show that the youngest consumers buy the cheapest cars. Consumers aged 25-30
buy 10% more expensive vehicles on average, and the premium is about 18–20% for other
age groups. Consumers aged 70 and above buy less expensive cars than consumers in all age
categories between 30 and 70.
32
The vehicle cost is the dealer’s cost for the vehicle. Defined in our dataset at “[T]he retailer’s ‘net’ or
‘dead’ cost for the vehicle and for both factory and dealer installed accessories contracted for at the time of
sale.” This includes transportation cost, and also includes a measure of installed accessories.
26
In Column 2 of Table 9 we estimate the same relationship, but within vehicle models.
Unsurprisingly, we find that age related differences are considerably smaller. While the very
youngest and oldest consumers purchase less expensive trims (and options), the difference
between these consumers and the rest is at most one percent. In column 3, we estimate this
relationship within a vehicle model-trim. In this case, there is no clear relationship between
customer age and cost. Costs vary within model-trims only from adding aftermarket options
or warranties. We find that consumers under 30 are least likely to buy extra options or
warranties.
From column 1 we know that female consumers buy vehicles that are 12% cheaper on
average than males. Within models this difference disappears (see column 2). Results from
column 3 indicate that female consumers spend slightly more on options and warranties
than men. While this difference (0.3%) is statistically significant, it is almost economically
insignificant (at approximately $75 in additional costs).
In our regressions we include indicators for the timing of the transaction, and whether the
vehicle was leased or financed through the dealer. Column 1–3 show that consumers financing
buy cheaper models, but more expensive trims, or additional options. Consumers leasing
vehicles exhibit similar trends. Weekends see the sales of slightly cheaper vehicles on average,
however on controlling for make and trim, the effect is tiny. On average, slightly more
expensive vehicles are sold at the end of the month and the year, but again on controlling
for model and trim the effects are small.
In Table 10 we present our results from regressing vehicle cost on the joint distribution
of age and gender. We use fixed-effects as discussed for the previous table. Our omitted
category is now male consumers under the age of 25. The results in column 1 imply that
in every age category, female consumers buy cheaper vehicles on average than males of the
same age, showing that the earlier result on gender holds across age groups. Indeed the
male-female new car premium is a remarkably robust 13–16% in almost every age category.
Young women buy the cheapest cars of any gender-age group, while 35 to 65 year old men
buy the most expensive cars.
The large coefficients that we observe in column 1 of Table 10 disappears in column 2,
on controlling for the model purchased. Thus, there is no systematic relationship between
customer demographics and the choice of trims within a given model, just as we saw in
Table 9. In column 3, the coefficients on each demographic group are positive and statistically
significant, indicating that men under the age of 25 spend the least on aftermarket options
and warranties. Examining the coefficients suggests that, beyond the age of 40, women spend
slightly more on these add-ons than men of the same age, but the differences are small.
Our results so far indicate that consumer demographics are systematically related to the
purchase price of the vehicle, but that this is almost entirely due to choices regarding which
make and model to purchase. Conditional on the model, demographics do not have a clear
relationship with consumers’ choice of vehicle trims or other options. We now turn to the
question of how consumer demographics relate to dealer profits.
A.2
Robustness
This section contains additional tables to establish the robustness of our main results.
In Table 11 we present regressions for specific subsets of our data. We start by examining
27
Table 10: Regression
Prov
Age < 25 Female
-0.157a (0.004)
Age 25-30 Male
0.077a (0.003)
Age 25-30 Female
-0.054a (0.003)
Age 30-35 Male
0.140a (0.003)
Age 30-35 Female
0.027a (0.004)
Age 35-40 Male
0.174a (0.003)
Age 35-40 Female
0.074a (0.003)
Age 40-45 Male
0.182a (0.003)
Age 40-45 Female
0.065a (0.003)
Age 45-50 Male
0.169a (0.003)
Age 45-50 Female
0.049a (0.003)
Age 50-55 Male
0.160a (0.003)
Age 50-55 Female
0.043a (0.003)
Age 55-60 Male
0.169a (0.003)
Age 55-60 Female
0.039a (0.004)
Age 60-65 Male
0.173a (0.003)
Age 60-65 Female
0.029a (0.004)
Age 65-70 Male
0.146a (0.004)
Age 65-70 Female
0.014a (0.004)
Age > 70 Male
0.099a (0.003)
Age > 70 Female
-0.025a (0.004)
Financed Indicator -0.034a (0.001)
Leased Indicator
0.031a (0.001)
Sat or Sun FE
-0.005a (0.001)
End of Month FE
0.015a (0.001)
End of Year FE
0.010b (0.004)
Constant
10.005a (0.008)
2
R
0.171
of Log(Vehicle Cost)
Model-Prov
Model-Trim-Prov
a
-0.007 (0.001) 0.003a (0.001)
0.001 (0.001) 0.005a (0.001)
-0.004a (0.001) 0.005a (0.001)
0.003a (0.001) 0.008a (0.001)
-0.002c (0.001) 0.006a (0.001)
0.004a (0.001) 0.009a (0.001)
0.003b (0.001) 0.010a (0.001)
0.004a (0.001) 0.009a (0.001)
0.003b (0.001) 0.010a (0.001)
0.004a (0.001) 0.008a (0.001)
0.005a (0.001) 0.011a (0.001)
0.003b (0.001) 0.008a (0.001)
0.007a (0.001) 0.014a (0.001)
0.005a (0.001) 0.009a (0.001)
0.007a (0.001) 0.015a (0.001)
0.005a (0.001) 0.010a (0.001)
0.006a (0.001) 0.015a (0.001)
-0.001 (0.001) 0.008a (0.001)
0.004b (0.002) 0.016a (0.001)
-0.006a (0.001) 0.007a (0.001)
-0.001 (0.002) 0.015a (0.001)
0.019a (0.001) 0.022a (0.000)
0.027a (0.001) 0.029a (0.000)
-0.001a (0.001) -0.001b (0.000)
-0.001a (0.000) -0.001a (0.000)
-0.001 (0.002) -0.001 (0.001)
10.226a (0.003) 10.199a (0.002)
0.896
0.935
c
p < 0.1, b p < 0.05, a p < 0.01. Fixed-effects specified as Col. 1: Province; Col. 2:
Model-Year-Province; Col. 3: Model-Year-Trim-Province. All regressions include yearmonth FEs. N=510866.
28
particular segments— Compact cars in column 1 and SUVs and Vans in column 2. These
segments are disproportionately associated with particular age groups, as we saw in Section 2;
younger consumers are more likely to purchase small cars, while middle-aged consumers,
especially those that are married, are more likely to purchase family cars such as SUVs
and minivans. Finally, in column 3 we restrict the sample to high-selling models, which we
define as those with at least 5000 in sales. All regressions include fixed-effects for modelyear-province-trim as well as for year-month.
In Table 15 we show standard errors, clustered in a number of different ways, for three
of the robustness checks employed in Section 4.1.
A.3
Employment Rates and Education
In Table 17 we present a comparison of female employment rates across cohorts. Our data for
this table is from the Organization of Economic Cooperation and Development, StatExtracts.
Each row presents the employment rate of a cohort across 5 year intervals. Employment rates
change as people age. A comparison of employment rates in the same year says little about
their employment outcomes in the years when they were partipating in the labor force. A
sensible comparison is at the same age. So in comparing these cohorts, we compare the
employment rate for those aged 25-29 in year 2006 (employment rate 77%), to those aged
25-29 in year 1976 (employment rate 52%). In other words, females aged 25-29 in 2006 were
much more likely to be employed than those aged 55-59 (in 2006) when they were also 2529 (comparing employment outcomes of those born between year 1976-80, with those born
between 1946-50).
In Table 18 we present a corresponding comparison of male employment rates across
cohorts. Our data for this table is also from the Organization of Economic Cooperation and
Development, StatExtracts. Each row presents the employment rate of a cohort across 5
year intervals. Those aged 25-29 in year 2006 (employment rate 84%) are compared with
those aged 25-29 in year 1976 (employment rate 88%). In other words, men aged 25-29 in
2006 were less likely to be employed than those aged 55-59 (in 2006) when they also were
25-29 (comparing employment outcomes of those born between year 1976-80, with those
born between 1946-50).
In Table 19 we present a comparison of male and female education rates rates across
age groups. Our data for this table is from the Organization of Economic Cooperation
and Development, StatExtracts. Each row presents the percent of population with tertiary
education from the year 2000–2011. With education, we do not need to compare cohorts
anymore, comparing female educational attainment across age groups is enough. In Canada,
young, aged 25-34, females are much more likely to have completed tertiary education than
older, aged 55-64, females (for the year 2006, we compare 62% to 38%). Given we are using
male customers as a control group in our study, it is also relevant to compare educational
attainment between women to men. The third column of the table illustrates the difference
betweeen the percent of population completing tertiary education between females and males.
If we look over all years, we find that younger (aged 25-34) females are more likely to have
completed tertiary education than men (the percent of females with tertiary education is
10-16% more than males). In contrast, older, aged 55-64, females are also more likely to
have completed tertiary education than males but the difference was significantly smaller
29
Table 11: Regression of Dealer Margin
Compacts
SUVs+Vans
High-selling
Age < 25 Female
-76.5a (13.4) -162.4a (48.4) -107.1a
(16.5)
a
b
Age 25-30 Male
3.2
(14.2) -164.9 (40.5) -37.6
(16.0)
a
a
a
Age 25-30 Female
-40.8
(13.4) -201.6 (40.7) -78.2
(16.0)
Age 30-35 Male
29.7b
(14.4) -182.0a (38.4) -25.7c
(15.6)
a
a
Age 30-35 Female
-8.0
(14.2) -164.6 (39.2) -48.2
(16.3)
Age 35-40 Male
24.1c
(14.5) -170.3a (38.0)
-12.5
(15.3)
Age 35-40 Female
28.4b
(14.5) -135.8a (38.9)
-18.2
(16.3)
a
a
Age 40-45 Male
42.5
(14.2) -139.2 (37.8)
-24.5
(15.1)
Age 40-45 Female
49.9a
(13.7) -95.7b (38.7)
13.8
(15.6)
b
a
Age 45-50 Male
32.7
(13.7) -112.6 (37.8)
-4.0
(14.9)
a
b
Age 45-50 Female
76.4
(13.3) -59.2 (38.8)
35.6
(15.3)
Age 50-55 Male
49.1a
(14.0) -85.0b (38.2)
12.6
(15.1)
a
a
Age 50-55 Female
96.8
(13.5) -41.9 (39.5) 74.2
(15.5)
Age 55-60 Male
76.3a
(14.7) -106.9a (38.6)
24.9
(15.6)
a
a
Age 55-60 Female
102.8
(14.2) -19.7 (40.9) 77.0
(16.3)
a
b
c
Age 60-65 Male
85.2
(16.1) -100.8 (39.6)
30.6
(16.7)
Age 60-65 Female
97.6a
(15.5)
-0.3
(43.7) 85.3a
(17.9)
a
b
a
Age 65-70 Male
110.8
(17.7) -102.7 (41.0) 56.5
(18.0)
Age 65-70 Female
109.7a (17.6) -11.9 (48.5) 112.6a
(20.2)
a
a
Age > 70 Male
123.3
(15.5) -26.9 (40.4) 101.0
(16.6)
a
a
Age > 70 Female
136.6
(16.3)
51.6
(50.9) 153.0
(19.0)
Financed Indicator 154.4a
(6.9) 216.4a (12.1) 149.1a
(7.3)
a
a
a
Leased Indicator
172.2
(6.9) 244.1
(12.0) 141.9
(7.2)
Sat or Sun FE
12.6c
(7.3)
28.6a (10.8) 23.8a
(7.2)
a
a
a
End of Month FE
-42.0
(4.9) -71.2
(8.3)
-61.2
(5.1)
a
a
a
End of Year FE
-108.9 (21.6) -109.6 (30.6) -90.4
(21.5)
Constant
1642.3a (36.7) 961.4a (73.4) 1900.2a
(37.0)
2
R
0.170
0.266
0.154
Obs
207125
142459
250588
c
p < 0.1, b p < 0.05, a p < 0.01.
Model*Year*Trim*Province FEs.
30
All regressions include year-month and
Table 12: Regression of Dealer Margin
Pos. Margin
Domestic
Foreign
Age < 25 Female
-106.4a (12.7) -122.5a (23.3) -96.4a
(15.8)
b
b
Age 25-30 Male
-30.3
(11.8) -42.8
(20.7)
-6.2
(15.2)
a
a
a
Age 25-30 Female
-77.8
(12.1) -76.5
(22.4) -67.8
(15.0)
Age 30-35 Male
-53.5a (11.5) -86.5a (19.8)
-8.9
(14.9)
Age 30-35 Female
-64.0a (12.2) -74.4a (22.0) -38.2b
(15.3)
Age 35-40 Male
-52.5a (11.3) -73.7a (19.3)
-11.2
(14.8)
Age 35-40 Female
-34.5a (12.1) -46.4b (21.3)
-0.9
(15.5)
a
a
Age 40-45 Male
-36.2
(11.1) -80.3
(18.8)
23.0
(14.6)
Age 40-45 Female
-5.8
(11.6)
-9.6
(20.4)
31.9b
(14.9)
Age 45-50 Male
-22.9b (11.0) -59.5a (18.7) 37.8a
(14.4)
b
a
Age 45-50 Female
24.7
(11.5)
16.3
(20.4) 66.8
(14.6)
Age 50-55 Male
1.5
(11.1) -39.3b (19.0) 54.0a
(14.6)
a
a
Age 50-55 Female
40.8
(11.7)
31.5
(21.1) 79.7
(14.7)
Age 55-60 Male
-3.1
(11.5) -57.5a (19.5) 60.1a
(15.1)
a
a
Age 55-60 Female
56.5
(12.3)
35.0
(22.3) 100.7
(15.5)
a
a
Age 60-65 Male
0.3
(12.2) -68.6
(20.6) 72.6
(16.1)
Age 60-65 Female
56.7a
(13.6)
12.1
(24.7) 100.8a
(17.0)
c
a
Age 65-70 Male
22.6
(13.2) -21.4 (22.1) 80.9
(17.7)
Age 65-70 Female
57.0a
(15.5)
18.4
(27.7) 115.8a
(19.4)
a
a
Age > 70 Male
43.4
(12.3)
10.0
(20.7) 130.0
(16.4)
a
a
a
Age > 70 Female
99.5
(14.6) 71.3
(25.7) 152.0
(18.5)
Financed Indicator 145.6a
(5.4) 269.3a (10.5) 133.5a
(6.3)
a
a
a
Leased Indicator
109.1
(5.3) 283.3
(10.4) 152.7
(6.2)
Sat or Sun FE
22.2a
(5.1)
40.0a
(8.7)
1.8
(6.7)
End of Month FE
-51.7a
(3.7) -86.6a
(6.4)
-34.2a
(4.8)
a
a
a
End of Year FE
-86.7
(15.3) -102.9 (24.9) -70.4
(20.3)
Constant
1396.5a (31.4) 901.9a (67.7) 2434.2a
(35.6)
2
R
0.290
0.184
0.371
Obs
467713
249262
261604
c
p < 0.1, b p < 0.05, a p < 0.01.
Model*Year*Trim*Province FEs.
31
All regressions include year-month and
Table 13: Regression of Dealer Margin
(1)
(2)
(3)
No Trade-Ins
No Leases
No Financing
Age < 25 Female
-116.2a (15.1) -97.1a (18.8) -131.7a
(18.4)
c
a
Age 25-30 Male
-26.7
(14.5)
4.0
(17.8) -55.9
(17.0)
Age 25-30 Female
-84.6a (14.8) -58.8a (18.3) -104.0a
(17.3)
b
a
Age 30-35 Male
-35.3
(14.1) -15.3 (17.4) -79.9
(16.4)
a
b
a
Age 30-35 Female
-59.1
(15.0) -44.7
(18.6) -78.3
(17.3)
(16.1)
Age 35-40 Male
-34.8b (13.9) -27.6 (17.1) -62.9a
Age 35-40 Female
-32.3b (15.0)
5.6
(18.4) -56.4a
(17.2)
c
Age 40-45 Male
-22.9
(13.6) -20.6 (16.6) -43.0a
(15.8)
c
Age 40-45 Female
25.8
(14.3)
21.8
(17.4) -10.6
(16.6)
Age 45-50 Male
-8.6
(13.4) -10.1 (16.3) -20.5
(15.7)
Age 45-50 Female
43.2a
(14.1) 63.5a (17.0)
14.8
(16.4)
Age 50-55 Male
7.8
(13.6)
-3.2
(16.4)
6.2
(15.9)
a
a
b
Age 50-55 Female
52.4
(14.4) 67.1
(17.3) 33.2
(16.7)
Age 55-60 Male
10.2
(14.2) -19.3 (16.8)
2.6
(16.4)
Age 55-60 Female
64.3a
(15.3) 71.8a (18.1) 52.0a
(17.4)
Age 60-65 Male
3.9
(15.3) -24.4 (17.8)
6.5
(17.1)
b
a
b
Age 60-65 Female
43.0
(17.2) 81.2
(19.7) 47.8
(18.9)
b
c
Age 65-70 Male
37.2
(17.1)
12.1
(19.1) 33.2
(18.3)
Age 65-70 Female
60.3a
(20.0) 71.1a (22.1) 56.6a
(21.1)
Age > 70 Male
56.6a
(16.0) 61.2a (17.6) 72.5a
(16.9)
a
Age > 70 Female
101.5
(19.0) 104.6a (20.7) 113.8a
(19.4)
Financed Indicator 222.4a
(7.4) 152.3a (6.2)
Leased Indicator
242.3a
(7.2)
212.8a
(5.8)
a
a
Sat or Sun FE
26.5
(6.5)
19.9
(7.4)
15.9b
(7.0)
a
a
a
End of Month FE
-62.8
(4.7) -55.6
(5.7) -60.6
(4.9)
End of Year FE
-96.9a (19.0) -70.7a (22.9) -101.1a
(20.0)
Constant
2361.3a (34.1) 967.3a (40.3) 583.3a
(49.8)
2
R
0.274
0.299
0.301
Obs
355952
273171
308711
c
p < 0.1, b p < 0.05, a p < 0.01.
Model*Year*Trim*Province FEs.
32
All regressions include year-month and
Table 14: Regression
(1)
Age < 25 Female
-114.6a (13.7)
Age 25-30 Male
-25.6b (12.8)
Age 25-30 Female
-81.2a (13.1)
Age 30-35 Male
-47.4a (12.4)
Age 30-35 Female
-60.5a (13.2)
Age 35-40 Male
-42.1a (12.2)
Age 35-40 Female
-24.6c (13.1)
Age 40-45 Male
-29.7b (12.0)
Age 40-45 Female
9.8
(12.6)
Age 45-50 Male
-11.6
(11.8)
a
Age 45-50 Female
40.8
(12.4)
Age 50-55 Male
6.0
(12.0)
Age 50-55 Female
54.2a
(12.7)
Age 55-60 Male
-0.6
(12.4)
Age 55-60 Female
68.8a
(13.3)
Age 60-65 Male
-1.5
(13.1)
a
Age 60-65 Female
60.0
(14.7)
Age 65-70 Male
26.9c
(14.2)
a
Age 65-70 Female
69.0
(16.7)
Age > 70 Male
63.7a
(13.3)
a
Age > 70 Female
114.5
(15.8)
a
Financed Indicator
185.7
(5.8)
Leased Indicator
205.0a
(5.7)
a
Sat or Sun FE
22.8
(5.5)
End of Month FE
-61.2a
(4.0)
a
End of Year FE
-94.1
(16.2)
a
Vehicle Cost (1000$)
-8.9
(0.6)
Log(Days Turned)
Constant
2688.8a (33.3)
Retailer FEs
No
R2
0.265
c
of Dealer
(2)
-115.2a
-26.6b
-82.1a
-49.4a
-61.8a
-44.2a
-26.8b
-32.0a
7.6
-13.8
38.2a
4.0
51.1a
-2.8
65.5a
-3.7
56.7a
25.3c
65.5a
62.3a
111.6a
180.6a
198.0a
22.9a
-60.7a
-94.1a
Margin
4.1a
2432.9a
No
0.265
(1.1)
(27.7)
(13.7)
(12.8)
(13.1)
(12.4)
(13.2)
(12.2)
(13.1)
(12.0)
(12.6)
(11.9)
(12.4)
(12.0)
(12.7)
(12.4)
(13.3)
(13.1)
(14.7)
(14.2)
(16.7)
(13.3)
(15.8)
(5.8)
(5.7)
(5.5)
(4.0)
(16.2)
(3)
-101.3a
-19.6
-67.3a
-43.5a
-53.4a
-33.0a
-16.3
-20.4c
17.2
-4.7
48.2a
13.9
64.4a
16.4
85.4a
10.9
77.8a
39.5a
80.3a
80.7a
131.8a
207.7a
236.0a
11.0b
-53.4a
-91.6a
3254.9a
Yes
0.341
(13.0)
(12.1)
(12.4)
(11.8)
(12.5)
(11.6)
(12.4)
(11.3)
(11.9)
(11.2)
(11.8)
(11.4)
(12.0)
(11.7)
(12.6)
(12.5)
(14.0)
(13.5)
(15.8)
(12.6)
(15.0)
(5.5)
(5.5)
(5.2)
(3.8)
(15.3)
(715.6)
p < 0.1, b p < 0.05, a p < 0.01. All regressions include year-month and
Model*Year*Trim*Province FEs. N=510866.
33
Table 15: Significance Tests for the Gender Gap
Age Group
Premium
P-value according to clustering:
No Leases
Under
25 to
65 to
Over
25
30
70
70
(Female-Male) None Dealer
-97.1 0.000 0.000
-62.8 0.000 0.000
59 0.006 0.006
43.4 0.017 0.009
Model-Year
0.000
0.000
0.007
0.012
Year-Month
0.000
0.000
0.006
0.005
Age Group
Compact cars
Premium
P-value according to clustering:
(Female-Male) None Dealer Model-Year Year-Month
Under 25
-75.3 0.000 0.000
0.000
0.000
25 to 30
-43 0.001 0.001
0.001
0.002
65 to 70
-6 0.762 0.772
0.787
0.782
Over 70
8.6 0.597 0.645
0.602
0.556
Age Group
High-selling
Premium
P-value according to clustering:
(Female-Male) None Dealer Model-Year Year-Month
Under 25
-105.9 0.000 0.000
0.000
0.000
25 to 30
-40.3 0.005 0.009
0.003
0.005
65 to 70
53.1 0.009 0.012
0.018
0.019
Over 70
48.5 0.006 0.008
0.011
0.000
(only 0-2%). The contrast is starker in the United States, where young females are more
likely to have completed tertiary education than males, but older women are less likely to
have completed tertiary education. However, in both cases, over time, female completion
rates are improving relative to males. Overall, female education achievement is improving
over time. This is true if we compare younger and older females, and but also true if we
compare females achievement relative to males over time.
A.4
Internet Use
In Table 20 we present the results of Statistics Canada’s internet use survey (held every two
years) from 2005-2012. Most individuals below 34 years of age (over 88% for all columns) use
the internet for personal use. Most individuals aged 65 and older (below 47.5% and as low
as 23.8%) report that they do not use the internet for personal use. This could be as older
people might be unfamiliar with newer technology, and have a higher cost of learning this
technology than the young. Irrespective of the reason for their low access to the internet,
this data supports the possibility that older consumers do not access the internet to research
their vehicles, and also do not use it to shop around for bargains. For this reason, older
individuals could pay a higher price for the same vehicles.
34
Table 16: Regression of Dealer Margin
(1)
(2)
Ontario
Quebec
Age < 25 Female
-80.8a (26.0) -124.5a (25.8)
Age 25-30 Male
-21.1
(24.0)
-9.5
(25.3)
Age 25-30 Female
-48.5b (24.0) -73.2a (24.8)
Age 30-35 Male
-7.4
(22.9)
-5.5
(24.9)
Age 30-35 Female
-13.7
(23.8) -75.2a (25.0)
Age 35-40 Male
4.5
(22.3) -12.7 (24.6)
Age 35-40 Female
31.6
(23.4) -32.8 (25.1)
Age 40-45 Male
33.3
(21.9) 39.9c (23.8)
Age 40-45 Female
69.5a
(22.7)
9.2
(23.8)
Age 45-50 Male
58.0a
(21.8) 48.3b (23.4)
Age 45-50 Female
100.6a (22.6)
35.9
(23.3)
b
a
Age 50-55 Male
56.2
(21.9) 83.8
(23.6)
Age 50-55 Female
130.4a (22.9) 57.3b (23.4)
Age 55-60 Male
52.7b
(22.2) 99.3a (24.3)
a
Age 55-60 Female
140.1
(23.6) 71.7a (24.2)
Age 60-65 Male
70.5a
(23.0) 57.3b (25.2)
Age 60-65 Female
151.5a (25.3)
34.7
(25.7)
Age 65-70 Male
86.3a
(24.2) 87.6a (26.8)
Age 65-70 Female
145.4a (27.8) 77.8a (28.4)
Age > 70 Male
143.5a (22.8) 89.3a (25.3)
Age > 70 Female
192.6a (25.7) 123.9a (27.9)
Financed Indicator 143.8a
(7.9) 263.3a (11.5)
a
Leased Indicator
322.4
(7.7) 167.2a (11.2)
Sat or Sun FE
1.3
(7.6)
-64.8c (37.2)
End of Month FE
-59.4a
(5.9) -39.8a
(7.3)
a
End of Year FE
-78.0
(23.4) -80.4b (34.0)
Constant
2124.4a (41.8) 281.6a (57.5)
2
R
0.254
0.240
Obs
196240
138730
c
p < 0.1, b p < 0.05, a p < 0.01. All regressions include
year-month and Model*Year*Trim*Province FEs.
35
Birth Year
1986-90
1981-85
1976-80
1971-75
1966-70
1961-65
1956-60
1951-55
1946-50
1941-45
1936-40
1931-35
Before 1930
Emp
Rate
2006
(Age)
Emp
Rate
2001
(Age)
Emp
Rate
1996
(Age)
Emp
Rate
1991
(Age)
Emp
Rate
1986
(Age)
Emp
Rate
1981
(Age)
Emp
Rate
1976
(Age)
47%
(15-19)
71%
(20-24)
77%
(25-29)
76%
(30-34)
77%
(35-39)
78%
(40-44)
79%
(45-49)
75%
(50-54)
59%
(55-59)
35%
(60-64)
12%
(65-69)
4%
(70-74)
1%
(75+)
44%
(15-19)
68%
(20-24)
75%
(25-29)
74%
(30-34)
75%
(35-39)
77%
(40-44)
76%
(45-49)
69%
(50-54)
50%
(55-59)
26%
(60-64)
7%
(65-69)
3%
(70-74)
39%
(15-19)
65%
(20-24)
70%
(25-29)
70%
(30-34)
70%
(35-39)
73%
(40-44)
70%
(45-49)
61%
(50-54)
45%
(55-59)
21%
(60-64)
7%
(65-69)
47%
(15-19)
68%
(20-24)
70%
(25-29)
69%
(30-34)
71%
(35-39)
73%
(40-44)
69%
(45-49)
59%
(50-54)
43%
(55-59)
22%
(60-64)
45%
(15-19)
68%
(20-24)
67%
(25-29)
66%
(30-34)
65%
(35-39)
67%
(40-44)
61%
(45-49)
51%
(50-54)
39%
(55-59)
46%
(15-19)
68%
(20-24)
62%
(25-29)
59%
(30-34)
60%
(35-39)
60%
(40-44)
55%
(45-49)
48%
(50-54)
41%
(15-19)
63%
(20-24)
52%
(25-29)
47%
(30-34)
50%
(35-39)
50%
(40-44)
48%
(45-49)
Source: OECD StatExtracts—accessed Nov. 2013
Table 17: Female Employment Rate by Cohort (Canada)
36
Birth Year
1986-90
1981-85
1976-80
1971-75
1966-70
1961-65
1956-60
1951-55
1946-50
1941-45
1936-40
1931-35
1930 and earlier
Emp
Rate
2006
(Age)
Emp
Rate
2001
(Age)
Emp
Rate
1996
(Age)
Emp
Rate
1991
(Age)
Emp
Rate
1986
(Age)
Emp
Rate
1981
(Age)
Emp
Rate
1976
(Age)
43%
(15-19)
72%
(20-24)
84%
(25-29)
88%
(30-34)
88%
(35-39)
87%
(40-44)
87%
(45-49)
84%
(50-54)
72%
(55-59)
50%
(60-64)
22%
(65-69)
9%(7074)
5%
(75+)
43%
(15-19)
70%
(20-24)
83%
(25-29)
87%
(30-34)
87%
(35-39)
86%
(40-44)
86%
(45-49)
82%
(50-54)
68%
(55-59)
44%
(60-64)
15%
(65-69)
11%
(70-74)
39%
(15-19)
68%
(20-24)
80%
(25-29)
83%
(30-34)
84%
(35-39)
84%
(40-44)
84%
(45-49)
80%
(50-54)
66%
(55-59)
40%
(60-64)
17%
(65-69)
47%
(15-19)
67%
(20-24)
79%
(25-29)
84%
(30-34)
86%
(35-39)
86%
(40-44)
86%
(45-49)
82%
(50-54)
69%
(55-59)
44%
(60-64)
46%
(15-19)
72%
(20-24)
83%
(25-29)
87%
(30-34)
88%
(35-39)
89%
(40-44)
87%
(45-49)
84%
(50-54)
66%
(55-59)
49%
(15-19)
77%
(20-24)
87%
(25-29)
91%
(30-34)
92%
(35-39)
91%
(40-44)
90%
(45-49)
82%
(50-54)
45%
(15-19)
76%
(20-24)
88%
(25-29)
92%
(30-34)
92%
(35-39)
92%
(40-44)
87%
(45-49)
Source: OECD StatExtracts—accessed Nov. 2013
Table 18: Male Employment Rate by Cohort (Canada)
Age
25-34 yrs
2000
2006
2011
55-64 yrs
2000
2006
2011
Canada
United States
Male
Female
Difference
Male
Female
Difference
43
48
49
54
62
65
10
14
16
36
36
38
40
43
48
4
7
10
Male
Female
Difference
Male
Female
Difference
28
37
41
28
38
44
0
1
2
34
40
42
26
35
40
-8
-5
-2
Source: OECD StatExtracts—accessed Nov. 2013
Table 19: Tertiary Education Completion Rates (Canada, US)
37
Age group
Aged
Aged
Aged
Aged
34
35
55
65
years
to 54
to 64
years
and under
years
years
and over
2005
2007
2009
2010
2012
88.9
75
53.8
23.8
93.1
79.8
60.8
28.8
96.5
87.8
71.1
40.7
97.5
93
80.1
40.2
98.6
95.5
83.8
47.5
Source: Statistics Canada. Table 358-0124 - Canadian Internet use survey (%)
Table 20: Personal Internet Use by Age Group (% of population) in Canada
38