Li S.Z., Jain A.K. (eds.) Encyclopedia of Biometrics

Подождите немного. Документ загружается.

the i

th comparison pair, when i 6¼ i

. The degree of

correlation is summarized by r. This is not the ty pical

Pearson’s correlation coefﬁcient, rather it is the intra-

class correlation or here the intra-comparison pair cor -

relation. More details can be found in Schuckers [10 ].

Derivation of sample size calculations requires an

understanding of sampling variability in the estimated

error rate. Thus consider

V ½

g¼N

2

gð1 

gÞ N þ

i¼1

ðm

 1Þ

; ð3Þ

where N¼∑

i¼1

, and

g ¼ N

1

i¼1

j¼1

Fleiss et al. [11] has suggested the following moment-

based estimator for r

r ¼

gð1 

gÞ

i¼1

ðm

 1Þ

1

i¼1

j¼1

¼1

6¼j

ðY



gÞðY



gÞ:

Since

g is a linear combination, if n is large it is

reasonable to assume that the central limit theorem

holds, Serﬂing [12]. To produce a (1a)  100%

conﬁdence interval for g use

g  z

a=2

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

2

gð1 

gÞ



N þ

i¼1

ðm

 1Þ



;

ð5Þ

where z

a∕2

represents the 1a∕2th percentile of a

Gaussian distribution w ith mean 0 and variance 1.

Further, if m

¼ m for all i (3) simpliﬁes to

V ½

g¼ðnmÞ

1

gð1 

gÞ 1 þ rðm  1Þ½; ð6Þ

where N has been replaced by nm. This form will be

used to derive sample size calculations.

Turning from variance estimation to sample size

calculations, set the portion of (6) after the , the

margin of error, equal to some desired value B and

solve for n, the number of comparison pairs. Then the

following sample size calculation for making a 100

(1a)% CI with a speciﬁed margin of error of B is

obtained.

n ¼

1

gð1  gÞð1 þðm  1Þ rÞ

&’

; ð7Þ

where de is the next largest integar or ceiling function.

In order to create sample size calculations for a

▶ conﬁdence interval, it is necessary to specify,

among other things, the desired margin of error, B,

for the interval. As ment ioned above there are effec-

tively two sample sizes when dealing with performance

evaluation for biometric authentication devices. This

derivation here is for the number of comparison pairs,

n, that need to be tested and assume that the number of

decisions per individual is ﬁxed and known. This is

equivalent to assuming that m

¼ m for all i and that m

is known. In practice it will be possible to determine

different value s for n by varying m before proceeding

with a evaluation. As with all sample size calculations it

is important to note that speciﬁcation of a priori values

for the parameters in the model is necessary. In this

case that means it is necessary to estimate values for

g and r to be able to determine the number of indivi-

duals, n . Several strategies are reasonable and have

been discussed in the statistics literature for these a

priori speciﬁcations. See, e.g., Lohr [13]. Ideally, it

would be possible to make a pilot study of the process

under consideration and use actual process data to

estimate these quantities. Alternatively, it may be pos-

sible to use estimates from other studies perha ps

done under similar circumstances or with similar

devices. The last possibility is to approximate based

upon prior knowledge without data. Regardless of the

method used it is important to recognize that n is

a function of a, B, m, g and r. n varies directly with g

and r and inversely with a, m and B. Thus, a con-

servative approach to estimation of these quantities

would overestimate g and r and underestimate m.

This w ill produce a value for n that is likely to be larger

than required. Table 1 illustrates the use of (7). It is also

worth noting that most studies of this type have a

signiﬁcant drop out rate of individuals as the data

collection progresses. Thus it is adviseable to plan

Test Sample and Size. Table 1 Illustration of the

use of (7)

a B g m r n

0.05 0.005 0.01 10 0.4 700

0.05 0.01 0.01 10 0.4 175

0.01 0.005 0.01 10 0.4 1,209

0.05 0.005 0.02 10 0.4 1,386

0.05 0.005 0.01 5 0.4 792

0.05 0.005 0.01 10 0.1 290

1330

Test Sample and Size

a collection process that assumes some attrition in the

number of comparison pairs to be selected. The values

of a and B are likely to be set by investigators or by

standards bodies rather than the performance of the

process under study.

Equation (7) is straightforward for calculation of

the number of comparison pairs that need to be tested

when g = FNMR. It is less so when interest centers on

g = FMR. This is because for FNMR the number of

comparison pairs translates to the number of indivi-

duals, while for FMR the number of comparison pair s

is not proportional to the number of individuals. If all

cross-comparisons are used to estimate FMR, then one

can replace n with n

∗

 1) in (7). In that case n

∗

will be the number of individuals that need to be

tested.

Sample Selection

Once the number of individuals to be selected is deter-

mined, another important step is to specify the target

population of individuals to whom statistical inference

will be mad e. Having done so, a sample would ideally

be drawn from that group. However, this is not possi-

ble often. The next course of action is to specify a

sample that is as demographically similar to the target

population as possible. The group of individuals that

will compose the sample is often referred to as the

▶ ‘‘volunteer crew’’ or simply the ‘‘sample crew’’,

Mansﬁeld and Wayman [5]. The more similar the

sample crew is to the target population the more prob-

able it will be that the estimates based upon the sample

crew w ill be applicable to the target population. Often

the sample crew is chosen to be a convenience sample,

Hahn and Meeker [1]. Methodology for best selecting

the sample crew is an open area of research in

biometrics.

One useful tool for extrapolation from estimates

based upon the ‘‘crew’’ is post-stratiﬁcation.

▶ Post-

stratiﬁcation is a statistical tool for weighting a sample

representation after the sample has been taken so that

resulting estimates reﬂect the known population. Sup-

pose that, there are H non-overlapping demographic

groups of interest, or strata, and n

individuals have

been sampled from among the N

total individuals in

each strata. Further suppose that estimates of the error

rate,

, from each of the strata are know n. Then a

poststratiﬁed estimate of the error rate is

h¼1

: ð8Þ

An estimate of the variability of the predicted error

rate is

V ½

¼

h¼1



V ½

; ð9Þ

where

V ½

 can be calculated using the equation

found above. A (1a)  100% poststratiﬁcation con-

ﬁdence interval for the process error rate can then be

made using

 z

a=2

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

V ½



: ð10Þ

As above, use of the Gaussian distribution here is

justiﬁed by the fact that the estimated error rate,

is a linear combination of random variables.

Summary

Testing and evaluation of biometric devices is a difﬁ-

cult unde rtaking. Two crucial elements of this process

are the selection of the number of individuals from

whom to collect data and the selection of those indi-

viduals. Determining the number of indiv iduals to test

can be calculated based on (7). To obtain the number

of individuals that need to be tested, some process

quantities need to be speciﬁed. These speciﬁcation

can be based on previous studies, pilot studies or on

qualiﬁed approximations. Selection of the ‘‘crew’’ for a

study is a difﬁcult process. Ideally a sample from the

target population is the best, but a demographically

similar ‘‘crew’’ is often more attainable. The inference

from a demo graphically similar crew can be improved

by the use of poststratiﬁcation.

Related Entries

▶ Inﬂuential Factors to Performance

▶ Performance Evaluation, Overview

▶ Performance Measures

▶ Performance Testing Methodology Standardization

Test Sample and Size

1331

References

1. Hahn, G.J., Meeker, W.Q.: Statistical Intervals: A Guide for

Practioners. Wiley, New York (1991)

2. Deming, W.E.: On probability as a basis for action. Am. Statist.

29(4),146–152 (1975)

3. Wayman, J.L.: Conﬁdence interval and test size estimation for

biometric data. In: National Biometrics Test Center, Collected

Works 1997–2000. http://www.engr.sjsu.edu/biometrics/nbtccw.

pdf, pp. 89–99 (2000)

4. Wayman, J.L.: Conﬁdence interval and test size estimation for

biometric data. In: Proceedings of IEEE AutoID ’99, pp. 177–184

(1999)

5. Mansﬁeld, T., Wayman, J.L.: Best practices in testing and report-

ing performance of biometric devices on the web at http://www.

cesg.gov.uk/site/ ast/biometrics/media/BestPractice.pdf (2002)

6. Louis, T.A.: Conﬁdence intervals for a binomial parameter after

observing no successes. Am. Statist. 35(3), 154 (1981)

7. Jovanovic, B.D., Levy, P.S.: A look at the rule of three. Am.

Statist. 51(2), 137–139 (1997)

8. Doddington, G.R., Przybocki, M.A., Martin, A.F., Reynolds,

D.A.: The NIST speaker recognition evaluation: overview meth-

odology, systems, results, perspective. Speech Commun. 31

(2–3), 225–254 (2000)

9. Schuckers, M.E., Sheldon, E., Hartson, H.: When enough is

enough: early stopping of biometrics error rate testing. In:

Proceedings of the IEEE Workshop on Automatic Identiﬁcation

Advanced Technologies (AutoID) (2007)

10. Schuckers, M.E.: Estimation and sample size calculations for

correlated binary error rates of biometric identiﬁcation rates.

In: Proceedings of the American Statistical Association:

Biometrics Section [CD-ROM], Alexandria, VA, American Sta-

tistical Association (2003)

11. Fleiss, J.L., Levin, B., Paik, M.C.: Statistical Methods for Rates

and Proportions. Wiley, New York (2003)

12. Serﬂing, R.J.: Contributions to central limit theory for depen-

dent variables. Ann. Math. Stat. 39(4), 1158–1175 (1968)

13. Lohr, S.L.: Sampling: Design and Analysis. Duxbury Press (1999)

Text-Dependent

Speaker recognition systems are said to be text-

dependent when the sam e piece of text is used for the

enrolment and for the subsequent authentication

sessions. A major drawback of text-dependent systems

lies in the replay attacks that can be performed easily

with a simple recording device.

▶ Remote Authentication

▶ Speaker Recognition, Overview

▶ Speaker Recognition, Standardization

Text-Independent

Speaker recognition systems are said to be text-indepen-

dent when there is no constraint on the text spoken b y

the user . Very convenient to use, such systems are

although quite sensitive to replay attacks as any reco rding

of the user’s voice can be used to break into the system.

▶ Remote Authentication

▶ Speaker Recognition, Overview

▶ Speaker Recognition, Standardization

Text-Prompted

Such speaker recognition systems prompt the user to

repeat a randomly chosen sequence of words. The

system ﬁrst performs speech recognition to verify the

expected sequence of words, then the veriﬁcation takes

place. These systems achieve a good level of security by

preventing replay attacks.

▶ Remote Authentication

▶ Speaker Recognition, Overview

Text-to-Speech (TTS)

▶ Voice Sample Synthesis

Text-to-Speech (TTS) Synthesis

TTS synthesis is the process of converting human read-

able text input into audible speech output. It involves

several steps including text and linguistic analyses,

letter-to-sound conversion for pronouncing each word

in context, and, ﬁnally, speech synthesis to create the

speech waveform.

▶ Voice Sample Synthesis

1332

Text-Dependent

TFIR

The Total Frustrated Internal Reﬂection is one of the

physical principles on which is based a contact-based

optical sensor.

▶ Fingerprint, Palmprint, Handprint and Soleprint

Sensor

Thermal Biometrics

Thermal biometrics uses thermal imaging to support

human identiﬁcation. This is an alternative imaging

modality which can reveal different (image) features

for identiﬁcation purposes, and it requires sensor sys-

tems which are more specialized than conventional

video.

▶ Ear Biometrics

Thermogram

The heat pattern emitted from an object.

▶ Face Recognition, Thermal

Third Level Detail

This refers to the minute details of the ridges and deals

with the shape and relative position of the pores, the

edge shape of the ridges, and the alignment of individ-

ual ridges.

Third level detail can be the same, and if so, may

add up to the value and signiﬁcance of each individual

point and the total. However, if different or absent, it

does not prevent identiﬁcation because it cannot be

expected to reproduce the same in the latent and the

inked print due to its minute detail, its three

dimensional properties (of the source) and less ideal

conditions during printing.

▶ Fingerprint Matching, Manual

Threshold Limit Value (TLV)

TLV

is a trademark of the ACGIH. TLVs

are similar

in concept to maximum permissible exposures (MPE)

and permissible exposure limits (PEL), though they are

distinct. The ACGIH (www.acgih.org) deﬁnes TLVs as

follows.

Threshold limit values (TLVs

) are determinati ons

made by a voluntary body of in dependent knowledge-

able individuals who represent the opinion of the sci-

entiﬁc community that exposure at or below the level

of the TLV

does not create an unreasonable risk of

disease or injury.

TLVs

are not standards. They are guidelines

designed for use by industrial hygienists in making deci-

sions regarding safe levels of exposure to various physical

agents found in the workplace. TLVs

are health-based

values established by committees that review existing

published and peer-reviewed literature in various sci-

entiﬁc disciplines (e.g., industrial hygiene, occupation-

al medicine, and epidemiology). Since TLVs

are

based solely on health factor s, there is no consideration

given to economic or technical feasibility.

▶ Iris on the Move™

▶ Iris Device

Thresholding

Thresholding is a simple grayscale image segmentation

method in which individual pixels are marked as ‘‘ob-

ject’’ pixels if their value is greater than some threshold

value (assuming an object to be brighter than the

background) and a s ‘‘background’’ pixels otherwise.

Typically, an object pixel is given a value of ‘‘1’’ while

a background pixel is given a value of ‘‘0.’’

▶ Hand Shape

▶ Image Segmentation

Thresholding

1333

Throughput

The number of end users that a biometric system can

process within a given time interval; for examp le, 1

min. Different applications have varying requirements

on throughput. For instance, for employee attendance

checking scenario, users might expect a throughput of

at least 30 considering the rush hour situation.

▶ Performance Evaluation, Overview

Time and Attendance

SAMIR TAMER

,STEPHEN J. E LLIOTT

Ingersoll Rand Security Technologies, Campbell, CA,

USA

Purdue University, West Lafayette, IN, USA

Definition

Time & Attendance (TA) describes the ﬁeld of tracking

the hours that employees spend on the job, and com-

pensating them appropriately. The time and atten-

dance calculations are non-trivial, and include shift

differentials, holiday pay, vacation and sick pay, ﬂex

time, minimum increments, time quantization as well

as multiple employee roles, where the pay structure is

different for each role.

Overview of Time and Attendance

Time & Attendance (TA) describes the ﬁeld of tracking

the hours that employees spend on the job, and com-

pensating them appropriately. While such a succinct

deﬁnition implies that the task is trivially simple, it is

indeed not. The TA industry encompasses hundreds of

competing companies focusing on one or more con-

stituent par ts of the ﬁeld, such as

1. Data capture

2. Payroll calculations

3. Payroll disbursement

This article discusses the following topics: complexities of

payroll calculation, data capture paradigms, a sampling

of modern

▶ TA terminals, driving factors for choosing

biometric terminals over non-biometric terminals, appli-

cation-speciﬁc tradeoffs, a case study, and factors con-

founding the adoption of biometrics for TA.

Payroll Calculations

The value of the Time & Attendance market is the

ability of companies to oversee their employees and

operations. Time & Attendance facilitates the compen-

sation of employees for time worked, and provides

insight into employee attendance and absence. The

beneﬁts of accurate time calculations include a correct

assessment of operational costs, improved regulatory

compliance, and better management. An ancillary

beneﬁt to accurate time and attendance is employee

morale – according to a report by IDC payroll accuracy

is one of small businesses’ biggest concerns [1].

The calculation of payroll is non-trivial. Many

countries have instituted laws standardizing the amount

of time an employer may reasonably expect certain

classes of employee to work. Any hours above and

beyond this reasonable limit are considered ‘‘overtime’’

and must be paid at a higher hourly rate. For example, in

the United States, the Fair Labor Standards Act of 1939

(FLSA) deﬁnes the standard work week as 40 h, and

requires employers to pay a 50% premium for any

hours worked beyond that limit.

Note that employees must be paid overtime if they

were ‘‘suffered or permitted’’ [2] to work those hours;

thus employees can sometimes choose to work over-

time (regardless of authorization or business need)

unless speciﬁcally prohibited or controlled by the em-

ployer. The Department of Labor states that ‘‘The

reason (employees work overtime) is immaterial.

The hours are work time and are compensable.’’

The issue of overtime is confounded by the fact that

different labor laws cover persons employed in differ-

ent countries, stat es, cities, etc. The Department of

Labor encourages businesses to contact local Wage

and Hour Division ofﬁces to understand their respon-

sibilities [3]. Small businesses end up becoming

experts in their local labor laws, and large multination-

al corporations become experts in the laws at each site

(including work-from-home sites) around the world.

The only effective alternative to becoming a labor law

expert is to outsource payroll calculations to a

specialized TA company.

1334

Throughput

One example of the differences seen in laws from

place to place involves the classiﬁcati on of certain

work-hours as standard or overtime. As mentioned

above, the Fair Labor Standards Act deﬁned a standard

work week as 40 h. However, in the US state of

California, a standard work day is deﬁned as 8 h, so

for certain jobs any time beyond 8 h in a single day is

considered overtime.

Other examples of calculation complexities

include:

1. Shift differentials (paying a premium for working

nights or weekends)

2. Holiday pay (paying a premium for working on

recognized ho lidays)

3. Vacation & sick-pay (paying even though an em-

ployee did not work)

4. Flex time (not paying overtime when employees

shift work hours between days)

5. Minimum increments (e.g., paying workers a min-

imum of 4 h for showing up at the worksite, even if

they are sent home after 10 min)

6. Time quantization (e.g., rounding time to the near-

est 15 min vs. the nearest minute)

7. Employees with multiple roles, who are paid differ-

ent rates for each role

While this is by no means a comprehensive list, is

does illustrate the point that calculating employee

compensation can be very complex. It’s easy to see

why some companies might want to outsource this

activity, freeing up time to concentrate on their core

business.

Data Capture

Regardless of who calculates employee compensation,

somehow the raw data must be collected. When col-

lecting data, companies have choices on how they

decide to automate and integrate. An immature time

and att endance system is paper-based, and relies on the

error-prone step of deciphering employees’ handwrit-

ing to record their hours-worked. The next step up

from this approach is a mechanical clock using time

cards, as shown in Fig. 1. An employee using a me-

chanical clock would typically:

1. Arrive at work

2. Pick his/her card out of a rack

3. Place it into a time clock t o stamp the current date/

time onto the card

4. Place the card back in the rack

Someone from the accounting department would then

typically collect all the time cards every week and type

the resulting numbers into a spreadsheet, calculate

each employee’s hours worked, and calculate their

compensation. There is an interface with payroll, but

such an interface requires manual workarounds.

As described in the previous section, there are

many different rules and complexities for time and

attendance systems. This complexity is sometimes

solved by the third level of automation where automa-

tion of the time and attendance system encompasses

the majority of workers (but not necessary all), and is

automatically fed into the payroll system. It is only

when full electronic time capture for all and a fully

integrated system are installed that workforce manage-

ment and visibility of the operation can be revealed.

These various complexities in automat ion and integra-

tion drive the selection of a Time & Attendance system.

Today, most small businesses are semi-automated,

and rely on employee vigilance to ensure that their

hours were recorded correctly. These businesses also

rely on the ability of the accounting department to

create workarounds when errors are discovered.

Time and Attendance. Figure 1 Typical punchcard based

time clock.

Time and Attendance

1335

Electronic data capture is inherently different, in

that the

▶ time clock generates electronic data when-

ever an employee clocks-in or clocks-out. TA terminals

using electronic data capture record a

▶ clock-in or

▶ clock-out transaction when the employee presents

an authorized credential to the clock. The credential

may be an ID number, proximity card, smartcard,

magnetic stripe card, etc.

Typical Time and Attendance Terminals

There are three main classes of electronic time & at-

tendance terminals – card or proximity based, ﬁnger-

print, and hand geometry.

Figure 2 shows a proximity card system for small

business (it supports up to 250 employees). The time

clock automatically calcu lates the total hours worked,

including overtime. Employees interact with the sys-

tem by bringi ng their proximity badge close to the

reader, and a large display and internal speaker provide

feedback to the employee. The system interfaces with

software on Windows™ compatible machines.

There are also biometric systems for small busi-

nesses – one example is the WaspTime Biometric

Time & Attendance system (Fig. 3) which uses ﬁnger-

prints to recognize the employee (it supports up to

1,500 employees). The system is very similar in that it

also provides clear audible and visual feedback to

employees, and interfaces to software on Windows™

compatible machines. However, the larger display and

numeric keypad support richer interactions with the

employee.

Another biometric modality that is used for time

and attendance is hand geometry. Figure 4 shows the

HandPunch 4000 from Ingersoll Rand. Similar to the

ﬁngerprint term inal above, it supports rich interac-

tions to thousands of employees (up to 3,498) with a

clear display, audible feedback, and a keypad. The

HandPunch also includes function keys and can

apply validation tables to complex data entered at the

terminal. Such data includes department transfers, tips

collected, job codes, and pay codes [4].

Figure 5 shows the GT-400 timeclock from

Ingersoll Rand, which exempliﬁes a recent trend in

biometric time clocks: hardware manufacturers are

Time and Attendance. Figure 2 Wasp Barcode RFID

reader.

Time and Attendance. Figure 3 Wasp Biometric Time

and Attendance System.

Time and Attendance. Figure 4 HandPunch 4000 from

Ingersoll Rand.

1336

Time and Attendance

providing programmable Linux-based terminals so

that specialized TA software companies can run their

applications directly on the data-collection term inal

(not only on a PC). Such terminals typically have

large displays, graphical user interfaces, ATM-style

keypads, and high processing power (one variant of

the GT-400 claims to support >100,000 users on the

terminal [5]).

Driving Factors for Choosing

Biometrics

As demonstrated in the previous section, there are many

commonalities between biometric and nonbiometric

terminals. In fact, the only true differentiator is that

biometric terminals are, well... biometric. Nonbio-

metric terminals require a physical credential (such as a

magnetic stripe card or smart card) or a nonphysical

credential (such as a PIN, employee number, or

password) to identify the employee. However, such

terminals cannot determine if the credential was pre-

sented by the authorized user or someone else. This leads

to the primary driving factor for companies’ choosing to

employ biometric Time & Attendance systems.

Payroll fraud, also called ‘‘buddy punching,’’ occurs

when an employee’s timecard is punched by someone

else in order to credit that employee with more hours

than were actually worked. ‘‘Buddy punching’’ typically

involves two employees that share identiﬁcation

mechanisms such as tokens, passwords, or smart cards

between them in order to receive pay for, or portion of

pay for that day. An American Payroll Association

study discussed in its January 2002 issue of PayTech

Magazine found that over 5% of payroll costs in the

United States are fraudulent. For many companies,

shrinking or eliminating this fraud saves more money

than is spent purchasing a biometric terminal.

Is buddy punching rampant in all companies?

Absolutely not; it is carried out by a minority of

employees, and only at some companies. Sites that

are most prone to buddy punching encompass unsu-

pervised employees and those with little management

oversight. A lack of oversight can occur when any of

the following conditions arise:

1. Employees work off-site or without supervision

2. Employees have ﬂexible or staggered work-

schedules

3. Employees who rise into supervisory roles feel un-

comfortable disciplining their former peers

4. The workforce has high turnover, making it difﬁ-

cult for superv isors to know who should be where

when

5. Supervisors are not present when employees clock-

in or clock-out

Even sites exhibiting none of these conditions often

choose to employ biometrics as an inexpensive, conve-

nient alternative to card-based systems. Card readers

generally cost less than biometric terminals, but t he up-

front costs of purchasing cards plus the annual cost of

replacing cards or worn-out readers can tip the balance

towards biometric-only or PIN + biometric systems.

The convenience of biometric terminals stems from the

fact that employees cannot lose or forget their biomet-

ric credential, where they can easily forget or misplace

card-based credentials. When that happens, time is

wasted while they track down a supervisor, spending

their own time and the supervisor’s time entering man-

ual overrides to clock-in and clock-out.

Application Specific Tradeoffs

The Time & Attendance market is different than other

markets. It has different customers, users, use proﬁles,

Time and Attendance. Figure 5 GT-400 from Ingersoll

Rand.

Time and Attendance

1337

and product demands than other markets. For this

reason, the data collection terminals focused on the

TA market have different attributes than similar term-

inals targeting other markets (e.g., Access Control).

Some of the most striking differences concern:

1. Location – TA terminals are generally placed in em-

ployee-only areas inside a company. Since customers

and investors generally don’t see these areas, there is

less need for the terminal to be beautiful or svelte.

2. User interface – Employees sometimes verify their

hours-worked or enter/extract other data from TA

terminals, requiring more buttons and larger

screens than AC terminals.

3. Vandal resistance – TA terminals must (surprisingly)

be more vandal-resistant than other terminals. This

is unintuitive since Access Control terminals are

often located on the outside of a building where

random people can attack the device with impunity.

However, random people rarely bother to attack a

small box hanging on the wall. Alternatively, employ-

ees who believe that a newly-installed TA system is

intrusive or onerous are sometimes irritated to the

point of physically damaging the system. Some even

use such damage to ‘‘prove’’ that the system is inher-

ently unsound and should be removed. Screwdrivers,

wire snips, pliers, paper clips, and even pens/pencils

can damage some TA terminals. Gum, dirt, glue, and

scratches effectively disable others.

4. Habituation – Employees typically clock-in and

clock-out every day. The y grow accustomed to

using the device over time, developing habits in

the way they use it. This process is called ‘‘habitua-

tion’’ and groups of users who have gone through it

are called ‘‘habituated’’ users. TA results in a highly-

habituated workforce.

5. Demographics – TA workforces are often demo-

graphically diverse, including employees of differ-

ing gender, age, ethnicity, size, and job function

(ofﬁce workers vs. manual laborers).

6. Error rates – TA data collection terminals control

employees’ paychecks, and thus must work for

every employee every time they attempt to use it.

At best, failure to do so results in wasted time

correcting the hours-worked. At worst it results in

incorrect paychecks or lawsuits. For this reason the

biometric tradeoff for TA systems is that the Failure

to Enroll Rate and False Reject Rate are every bit as

important as the False Accept Rate.

When assessing biometric reject rates, organizations

must insist on credible (independent) data collected

from a statistically-signiﬁcant numbe r of habituated

users with an appropriate demographic distribution

including the mix of ofﬁce workers and manual laborers

expected at the target site.

Case Study: McDonalds

This case study, excerpted from Ingersoll Rand’s web

site [6], is indicative of the types of applications served

by biometric time clocks.

In Venezuela, McDonald’s restaurants are cutting pay-

roll costs by up to 22% annually after incorporating

HandPunch biometric terminals to record time

and attendance.

Over 3,400 employees at 85 McDonald’s restaurants in

Venezuela have been enrolled with the HandPunch

over the past four years. On average, the system

generates over 7,500 transactions each day resulting

in over 2.5 million ‘‘punches’’ annually.

Students make up about 90% of the McDonald’s work-

force in Venezuela. They were frequently punching

one another into cover for exams or other school-

related events. McDonald’s needed to move to

biometrics to verify that the employee clocking in

was really that person.

Most supervisors at McDonald’s are promoted from

within and many ﬁnd it difﬁcult to impose rules

and restrictions on their fellow workers. The Hand-

Punch ensures that everyone is treated fairly (Fig. 6).

Confounding Issues

Since there are so many beneﬁts to using biometric TA

terminals, why don’t all companies use them? Because

they have down-sides to them too. The most common

issues cited when arguing against biome trics are:

1. Morale – Some employees ﬁnd it demeaning that

the employer doesn’t trust them to

▶ punch-in and

▶ punch-out correctly. If some workers/depart-

ments are required to use the TA terminal while

others are not (possibly because they are exempt

from FLSA), tensions can rise. One employee that

was dissatisﬁed stated ‘‘These systems are being

used...to reduce us to mere cogs in the machine’’...

‘‘It is as if we are working in some textile mill in the

1338

Time and Attendance

1820s and when the whistle blows we had better all

be sitting in front of our looms ready to go... As

civil servants, we produce incredibly creative,

thoughtful and substantive projects but this only

happens when we are treated with respect and dig-

nity, and these systems treat us with neither!’’ [7].

2. Privacy – The idea that ‘‘big brother’’ is watching is

unacceptable to some employees. They may link the

idea of biometric TA terminals to government bio-

metric programs they’ve heard about, including

criminal forensics, border control, and the search

for terrorists. Movies such as Minority Report in-

voke the image of uncontrolled biometric surveil-

lance tracking one’s every move [8].

3. Safety – Employees and unions often cite safety as

an argument against biometric time clocks. Com-

mon issues include ‘‘lasers’’ in the eye (a misinter-

pretation of the infrared LEDs used in many face

and iris systems) and microbes on the surfaces of

ﬁngerprint and hand geometry systems.

4. Restrictive work policies – Some TA systems incor-

porate rules & policies that employees ﬁnd oner-

ous, such as requiring that employees punch in/out

within 5 min of their scheduled start/ﬁnish times.

This has nothing to do with biometrics, but is often

blamed on the biometric terminal since it is the

most visible part of the system.

Sometimes complaints turn into grievances, and some

grievances must be solved through litigation or arbi-

tration. In the arbitration ruling of Canada Safeway

Ltd. and United Food and Commercial Workers Union,

Local 401, the arbitrator balanced the privacy rights of

employees to the business needs of the company. He

stated: ‘‘the more intrusive the impact on employee

privacy the greater the business rationale that must be

demonstrated. Conversely, if the intrusion on employ-

ee privacy is insubstantial, the concomitant level of

justiﬁcation also is lower.’’ He found that Hand Geom-

etry technology was minimally intrusive, and ruled

that employees could be required to use hand geome-

try time clock s [9].

In another arbitration ruling, IKO Industries Ltd.

and U.S.W.A., Loc. 8580, the arbitrator found that

capturing ﬁngerprints was not an ‘‘egregious disregard

of privacy rights’’ but was nonetheless an invasion of

privacy. IKO Industries was not able to show enough

business beneﬁt to justify that invasion, and the arbi-

trator ordered IKO to cease and desist their deploy-

ment of biometrics.

In addition to navigating the mineﬁeld of installing

a biometric TA system, employers must be mindful

of providing due process to employees during disciplin-

ary proceedings (should they become necessary). In a

non-biometric TA application in Tucson, Arizona, city

sanitation workers and union representatives said that

buddy-punching was allowed by supervisors, and was

something that had gone on for years in the city’s

Environmental Services Department [10]. The ﬁring

of two supervisors was overturned because the super-

visors weren’t given enough warnings, and both were

reinstated to their former positions and were awarded

back-pay.

Companies wishing to deploy biometrics to com-

bat buddy-punching must consider all of these con-

founding issues and precedents when choosing a

system. Engaging with unions and educating employ-

ees before introducing biometric terminals can allay

concerns before they escalate into grievances.

Summary

Time and Attendance is a complex ﬁeld covering pay-

roll calculations, labor law, and employee relations.

Historically, the data for payroll calculations has been

gathered by hand-writing time cards or using ‘‘punch’’

Time and Attendance. Figure 6 Biometric time clock at a

quick-serve restaurant.

Time and Attendance

1339