Cash, Check or Bank Card? The effects of transaction characteristics on the use of payment instruments

Cash, Check or Bank Card?
The effects of transaction characteristics on the use of
payment instruments?

David Bounie a and Abel François a,b

a Télécom Paris, Department of Economics and Social Sciences, Paris
b LAEP, Université Paris I – Panthéon-Sorbonne

June 2006

Exploiting a unique and original dataset of 11,945 payments made from March to May
2005, the paper estimates the determinants of the probability of a transaction being paid
by cash, check or bank card. Controlling for individual characteristics, the main results of
the paper are: 1) a differentiated effect of the transaction size regarding payment
instruments; 2) a specialization effect between payment instruments according to the type
of good and spending place; 3) a double supply-side effect due to a restriction of the
payment choice and to the organization of the payment process. For the first time ever,
detailed data on consumption and payment patterns make it possible to assess the role of
transaction characteristics in the use of payment instruments.

JEL Codes: G2, E4
Keywords: Payment instruments – money

? This research benefited from financial support from the Groupement des Cartes Bancaires « CB ». We thank
Yves Randoux and Olivier Mansart for their support and Marc Bourreau and Ludovic Lebart for helpful
suggestions and comments.
1

---

1 Introduction
Recent empirical investigations support the view that the substitution of electronic for paper-
based systems can reduce the social cost of a country’s payment system. Humphrey et al.,
(2003), for instance, note that “if a country moves from a wholly paper-based payment system
to close to an all electronic system, it may save 1% or more of its GDP annually once the
transaction costs are absorbed”. However, people are not necessarily prepared to adopt and
use electronic payment systems. The failures of several electronic purses in Europe, for
instance, are well documented (Hove, 2004) even if these payment instruments can be
theoretically less costly (Shy and al. 2002). Understanding the reasons why people adopt and
use electronic payment instruments can thus become a primary concern for banking and
monetary authorities. Why do people still prefer to use cash? Why are checks sometimes
preferred to bank cards to settle transactions? What are the factors that influence the use of a
payment instrument?

Even if payment is an everyday experience for all consumers, the understanding of the
use of alternative payment instruments has not, paradoxically, received much attention in the
literature (Boeschoten, 1998; Hancock and Humphrey, 2001). Theoretical works such as
Santomero (1974), Whitesell (1989), and Santomero and Seater (1996) provide some fruitful
analytical results on the choice of payment instruments but the lack or the poorness of data
(Hancock and Humphrey, 2001) have limited empirical verifications.

The purpose of this paper is precisely to explain the determinants of the use of payment
instruments. Following a survey undertaken from February to April 2005 on a representative
sample of the French population, we use a unique and original dataset to explain the
probability of a transaction being paid by cash, bank card or check, in terms of two sets of
explanatory variables: transaction and individual characteristics. We use 11,945 self-reported
transactions by individuals over an 8-day period in our econometric estimation based on a
multinomial logit model. Globally, we find: 1) strong evidence of a differentiated effect of the
size of transactions compared to payment instruments; 2) a specialization effect between
payment instruments according to the type of good and spending place; and 3) a double
supply-side effect due to a restriction of the payment choice and to the organization of the
payment process. For the first time ever, detailed data on consumption and payment patterns
enable us to assess the role of transaction characteristics in the use of payment instruments.
2

---

The rest of the paper is organized as follows. First, we present the related literature. We
then describe the methodology and comment on payment patterns. Thirdly, we present the
econometric analysis and fourthly, we discuss the estimation results. Finally, we conclude the
paper.
2 Related literature

Empirical investigations on the choice of payment instruments are relatively recent. Two
main groups of empirical studies may be commented on.

A first group of contributions attempts to analyse the influence of particular effects on
the use of payment instruments or the specific use of payment instruments such as electronic
payment systems.

For example, the works of Mantel (2000), Stavins (2001) and Hayashi and Klee (2003)
highlight the role of socio-demographic and technological factors on the propensity to use
electronic payment systems or the probability of them being used. Mantel (2000) proposes a
framework to describe why consumers pay bills electronically. Using US national survey
data, the author shows that wealth, personal preferences and certain demographic factors are
significant to explain the use of electronic bill payments. Stavins (2001) uses survey data from
US households in order to evaluate the effects of consumer characteristics on the probability
of using different electronic payment systems. The results show a strong effect of
demographic characteristics on consumers’ use of payment instruments. Lastly, Hayashi and
Klee (2003) use data on a sample of US consumers and primarily from users on the Internet in
order to test consumers’ propensity to adopt new technologies (debit cards, electronic bill
payment). The results indicate that consumers who use new technology or computers are
more likely to use electronic forms of payment and that payment choice depends on the
characteristics of the transaction such as the transaction value and the physical characteristics
of the points of sale (such as the absence of a cashier or the availability of self-service).

Other works such as Humphrey, Kim and Vale (2001) try to estimate a model of
payment choice (cash, debit card and check) from which own price, cross price and payment
substitution elasticities are derived. Using semi-annual panel data from Norway over the
period 1989-1995, the authors show that consumers are sensitive to prices.

Lastly, some works such as Rysman (2004) deal with problems related to network
effects in payment networks. Exploiting a unique dataset on the payment card industry, the
3

---

author shows the existence of single homing effects (concentration of spending on a single
payment network) and a positive feedback loop between consumer usage and merchant
acceptance.

A second group of contributions attempts to determine the influence of a whole group of
factors on the use of alternative payment instruments. From payment aggregate data collected
on fourteen developed countries over the period 1987-1993, Humphrey, Pulley and Vesala
(1996) analyze for instance the effects of economic and institutional determinants on the use
of five types of paper and electronic payment instruments. Several types of variable are
studied: price and income, equipment (cash holding), payment availability (Point of Sale
(POS), Automated Teller Machine (ATM)) and, finally, institutional variables (crime,
concentration of the banking system). The econometric results show in particular a limited
impact of own prices and a positive effect of the POS and ATM on the use of payment cards.
Carow and Staten (1999) attempt to explain consumers’ payment option to use cash, debit
cards, general purpose credit cards or gasoline credit cards. Using a survey of US gasoline
credit cardholders in 1992, the authors estimate the probability of using a payment instrument
in relation to mainly demographic and credit characteristics. The results show that consumers
are more likely to use cash when they have less education, lower incomes, are middle-aged
and hold more credit cards. Debit and credit card holders are less likely to use their gasoline
credit cards. Finally, Boeschoten (1998) seeks to explain payment behaviours of households.
The author uses the results of an ad hoc design panel survey over the 1990-1994 period, on
Dutch households. This study is the most complete concerning the covered population and the
explanatory variables used. In particular, the importance of transaction size is demonstrated:
indeed “the probability that an amount which is 100 per cent higher is paid cash is about 20
to 30 percentage points lower (...)” (Boeschoten 1998: 133).


A brief survey of the main contributions allows us to point out that individual,
technological or financial variables are mostly preferred as explanatory variables in
explaining the choice of payment instruments. However, several authors report that the choice
of payment instruments is derived from their use in transactions and, hence, that transaction
characteristics are predominant (Humphrey et al., 2001, Boeschoten 1998, Hayashi and al.
2003). Boeschoten (1998), for example, concludes that the transaction size is one of the most
important explanatory variables in choosing a payment instrument. However, apart from size,
transaction characteristics have received little attention in the literature mainly because of the
4

---

difficulty of collecting data on consumption and payment behaviours at multiple points of
sale. But there are numerous other variables that could be used to describe a transaction and
that could influence the use of a payment instrument such as the type of good purchased, the
type of spending place where the purchase occurred, the type of contact used for the
transaction (post mail, face-to-face, etc.) and supply-side constraints (limited choice of the
payment instrument). Because the role of transaction characteristics has not generally been
taken into account in previous empirical works, we seek to assess whether transaction
characteristics have a predominant role in explaining the use of payment instruments.
3 Data description
3.1 Survey design: a two-stage method
To study payment patterns, we undertook a survey using a two-stage method.

First, we administrated a survey from March to May 2005 on a representative sample of
1,447 French individuals of 18 years and older. The questionnaire aimed to collect
information related to banking and payment instrument equipment, and to obtain socio-
economic data on individuals (e.g. income, profession, etc).

Second, we asked each respondent to keep a diary in which they reported all information
related to purchases on a daily basis, for eight days. A purchase is characterized by six pieces
of information: the amount to be paid (size of transaction), the type of good or service
purchased, the type of spending place in which the good has been purchased, the type of
contact (face-to-face, Internet, telephone, etc.), the supply-side constraints and, finally, the
payment instrument used. Of the 1,447 respondents, 1,392 individuals completed the diary.
Overall, we have 16,692 transactions containing all information on transaction characteristics.
3.2 Payment patterns

Payment patterns are obviously dependent on the possession of payment instruments.
Globally, we note that 85% of individuals say they have a check-book and 83% a payment
card related to their main account. Among the range of available cards issued by banks, the
debit card is the most popular (85%). Only 5% of the French population says that it owns a
banking credit card. A very large majority of individuals (85%) has only one debit card.
5

---

Transactions are frequently made on Fridays and Saturdays; these two days are also the
most significant in value (the most frequent days without purchases are Sundays and
Mondays). The total amount of the transactions is 541,583 euros. On average, an individual
makes 12 purchases over eight days, i.e. just over one transaction per day.

The majority of transactions are under 15 euros (56% of all transactions). Purchases for
under 5 and 1 euro, respectively, account for 34% and 9% of all transactions. Interestingly,
purchases for over 15 euros account for only 4% of the total. This particular distribution of
transactions directly affects the choice of the payment instruments. Thus, we note that 97% of
the 16,692 transactions made by individuals were paid using cash (62%), bank card (21%) or
check (14%). Whereas cash is ahead of bank cards and checks in terms of the volume of
payments, the latter two payment instruments take the lead in terms of transaction value; they
respectively account for 36% and 33% of the total amount of purchases. Given the importance
of the use of these three payment instruments, we decided to focus us on them.

A more detailed analysis of the number of payments according to transaction size
enables us to identify the market shares of the three payment instruments (Figure 1).

[Insert Figure 1 around here]


We observe that cash payments decrease as the transaction size increases. Thus, cash is
more frequently used for low value transactions. For example, the cash market share for
transactions of under 5 euros is about 90%; this market share amounts to approximately 8%
and 2% for bank cards and checks. But the cash market share quickly decreases: when the
transaction size is around 23 euros, cash and bank card market shares are equal (35%).
Beyond that, bank card use increases up to values located around 60 euros, and significantly
decreases for values higher than 150 euros. Beyond 150 euros, payments by check are more
significant.

We also note in Figure 2 that the use of the payment instruments is not equally
distributed according to the different types of good.

[Insert Figure 2 around here]
6

---

Cash is mainly used to purchase goods or services such as “newspapers, tobacco and
lotteries”, “food and beverages”, “restaurant and hotel” and “culture and leisure”; checks are
preferably used in “health services” and bank cards in “transport” and “equipment and
personal care”. This specialization effect can be due to several factors. First, the size of the
transaction in each class of goods can differ and influence the use of specific payment
instruments; for instance, newspapers (low values payments) will preferably be settled by
cash; “health services” by check, etc. Second, the type of spending place in which the
purchases occurred may impact on the use of payment instruments. We observe a similar
effect in the distribution of payments by spending place (Figure 3).

[Insert Figure 3 around here]


Thus, we note that cash is mainly used in “small stores” whereas checks are more
present in “home services” and “public administration”. By contrast, the bank card is the
payment instrument preferably used in “department stores” and “supermarkets”. This
particular distribution of the use of the payment instruments that we call “specialization
effect” is probably due to supply-side constraints since “public administrations” and “home
services” do not generally accept bank cards. Furthermore, some spending places such as
department stores and supermarkets may facilitate or encourage the use of bank cards in order
to reduce the overall processing costs of payments.

Besides the role of the type of spending place on the use of payment instruments, the
type of contact may also have an effect (Figure 4).

[Insert Figure 4 around here]


In this figure we note that cash is used more in “face-to-face” transactions, checks in
“post mail” transactions and bank cards in “electronic transactions” (the Internet, automat
(vending machines, etc.) and telephone).
7

---

A further interesting observation relates to the number of payments where the choice of
the payment instrument was limited1 at the point of sale: approximately one transaction on ten
was settled in the framework of a limited choice. These payments are usually made during
face-to-face interactions (62%) or on automats (24%)2. On the whole, when the choice of the
payment instrument is limited at the point of sale, cash and bank cards are the most frequently
used instruments (58% and 23%).
4 Econometric estimations
In this part we assess the main determinants of the use of cash, checks and bank cards during
transactions.
4.1 Method

Initially, we have 16,692 transactions made by 1,392 individuals who completed the
diary. Three elements reduce the sample we use in the econometric analysis. First, transaction
and individual characteristics must be declared. Second, we exclude twenty-six transactions of
over 1,000 euros3. Finally, we are exclusively interested in transactions paid by cash, check or
bank card (debit and credit)4. Consequently, we exclude people who do not hold bank cards or
check accounts. We finally have a total of 11,945 observations available, for 1,035
individuals.

Table 1 below illustrates a high number of cash payments in volume (59.5% of all
transactions are paid in cash) whereas bank card payments dominate in value (44.4% of all the
payments).

[Insert Table 1 around here]

1 The choice of the payment instrument may be limited at the point of sale when for instance merchants do not
accept bank cards or when automats accept only coins, etc.
2 Despite the low number of transactions on the Internet (0.4% of all transactions), 57% of these transactions
were settled in the frame of a limited choice.
3 We note that these transactions are not very frequent over the observed period and therefore not
representative of consumer purchases. We then exclude these outlier purchases from our analysis.
4 Only 3% of all transactions were settled using other payment instruments (electronic purse, etc.).
8

---

We wish to estimate the choice of a payment instrument j made by an individual i during
a transaction k. The model rests on the utility associated with each choice. More precisely, the
utility can be explained by observable characteristics (
j
V ) and by unobservable
ik
characteristics ( j
? ):
ik
j
j
j
j
j
j
U = V + ? = ? X + ? Z + ?
ik
ik
ik
k
i
ik

The observable characteristics consist of transaction characteristics ( X ) and individual
k
characteristics ( Z ). We do not directly observe the utility of an individual i; we observe the
i
choice of a payment instrument j
I (with j = cash, check or bank card) for a transaction k. We
ik
can then estimate the probability that a consumer i chooses a payment instrument j during a
transaction k, i.e.:
Pr( j
I ) = Pr( j
k
U > U
k
? ? j) .
ik
ik
ik

To estimate the choice of a payment instrument we use a multinomial logit model.
Given the method of observation and the fact that the surveyed consumers do not make the
same number of transactions during the week, we have to control the impact of individual
variables. Traditionally with survey data, consumer characteristics are similar for all
transactions made by each individual. Consequently, we have to check whether the error term
is not correlated with unobservable effects for each consumer (defined as clustering). To do
so, we use the cluster-robust variance matrix which allows for adjustment of the standard
errors of the estimators (Cameron and Trivedi, 2005).
4.2 Explanatory variables
Two sets of explanatory variables related to transactions and individuals are used in the
regression.

Transactions are described by six characteristics.

The first is the size of the transaction which is indicated by its price and its square value.
Following our previous statistic descriptions, the impact of the transaction size on the
probability is assumed to be non-linear. More accurately, we anticipate that the marginal
effect is decreasing with the amount: the negative variation of the probability of using cash
9

---

and the positive variation of the probability of using a check or bank card should be
decreasing with transaction size.

The second and third variables are the type of good and the type of spending place
where the good is purchased. We capture these characteristics by dummy variables, each of
which describes a category of good or spending place (see Figure 2 and Figure 3). The
expected effects of these variables are unknown even if we anticipate a specialization of the
use of payment instruments according to the spending places (specialization effect). Indeed, as
we previously noted, some spending places encourage (or discourage) consumers to use
certain payment instruments. For instance, department stores and supermarkets in France have
equipped their point of sale with devices that facilitate payment by bank card. In small stores
there is often a single cash register where people queue to be served; consequently, even if a
small store accepts bank cards, people do not necessarily use them as they prefer to avoid
complaints from people waiting in the queue. This organization of the payment process within
the spending place obviously modifies the different costs of using alternative payment
instruments. In line with the theoretical argument of Whitesell (1989) who claimed that a
fixed transaction cost could consist of “extra time spent in a credit or check payment queue
versus a currency-only queue”, we can anticipate a specialization effect due to an (in)decrease
of the fixed costs per transaction.

The fourth variable is the type of contact measured by six dummies: face-to-face,
telephone, post mail, the Internet, automats and other types (see Figure 4). Since face–to-face
is excluded, we expect negative signs of the dummy coefficients concerning the probability of
using cash. The impacts on the other payment instruments are a priori indeterminate.

The fifth variable is the day of the week. This variable tries to take a temporal
dimension into account.

The sixth and last transaction characteristic focuses on supply-side constraints that limit
the choice of payment instruments. For instance, people cannot choose their favourite
payment instrument on the Internet, on automats or in retailers that do not accept all the
payment instruments. In order to capture supply-side constraints, we set up a dummy variable
that takes either the value one, if the choice is not limited, otherwise zero.

Finally, we use individual characteristics as control variables such as age, gender, level
of education, personal income and cash profession (Profcash). The cash profession is a
dummy variable which takes either the value 1, if individuals have an occupation in which
10

---