You are on page 1of 16

4.

3C Personnel selection - Group 5


Names:
● Sterre van Beek
● Luca Haaks
● Dejan Agovic
● Zoe Sou

Problem 1 - From organizational goals to operational predictors


Organizational Conceptional Operational criteria(be Conceptual Operational predictors(be
goals Criteria specific, percentage) predictors specific)

Patient care Social skills - Measuring successful - Extraversion - Filling in the Big Five
(pleasant/comforting) - Languages Questionnaire, the predictor will
conversation with a patient known be the score on extraversion in
by asking them afterwards the questionnaire
- Nurse should be able to - Language test that measures
speak Dutch and English Dutch and English language
properly skills

Provide Specific - Patients successfully - Relevant - Successful interaction with a


treatment knowledge in treated, this is measured by education difficult patient (roleplay)
nursing and seeing how many patients completed - Evidence like CV, jobs related
they related to to nursing, years worked as a
medical field
fed/bathed/dressed/groomed/ nursing nurse
moved/patients changed - work - Recommendation letters
linens in a day experience in
- And measure mistakes in the medical
the acts we mentioned above field

Efficiency - Planning - A supervisor measures how - The nurse - A roleplay where the nurse
- Using well the nurse works with knows how to works with a PDA with the apps
technology the PDA that contains apps work with the related to nursing installed that
(robotic are needed for the nursing job
for scheduling (supervisor apps on the
coworker) (the performance is measured by
- multitasking ratings) PDA related to supervisor ratings during the
-How many patients the nursing roleplay)
nurse successfully - -Big five questionnaire
fed/bathed/dressed/groomed/ Conscientiousn → score on conscientiousness
moved patients/changed ess
and neuroticism
-Low
linens in a day (measure of - Test knowledge about working
neuroticism
planning and multitasking) with a schedule. That is assessed
by a supervisor.- A roleplay
where the nurse has to perform
multiple tasks simultaneously.
The nurse is assessed on her
performance by a supervisor;
supervisor rating).
Problem 2 - There just are too many who want to study medicine
After their high-school exam, 400 pupils submit an application for the university study of
Medicine. Unfortunately for them, there is a numerus fixus of 150 positions at the university, and
therefore only 150 out of the 400 can be admitted.

The university can choose one out of several possible ways of selecting those 150 pupils. It will
choose one out of the following:
1. First, the university may decide to use a weighted version of random selection, where a
student's chance to be selected depends on the category of high-school exam result the
student is in.
2. Second, the university may want to use merit selection: top-down selection of the best
pupils according to their high-school exam results.
3. Third, the university may select the pupils at random, irrespective of their high-school
exam results, simply by drawing of lots.
4. Fourth, the university may use a quasi-random basis of choice, such as first come first
chosen.

Question:
1. Each choice will have a different outcome. Fill out the following table:
2. What are the numbers of admitted pupils in each category (I, II, III, IV) when weighted
random selection is used?
a. Answer: See table admitted pupils
3. What are the chances and numbers of admitted pupils per category when optimal merit
selection is used?
a. Answer: See table (chances and number of admitted pupils).
4. idem, but now for random selection?
a. Answer: See table (chances and number of admitted pupils)
5. Try to estimate the chances and numbers of admitted pupils per category under the
condition that the university uses the principle of “first come, first chosen”. Which
assumption do you need to make in your estimations? How strong/ weak is this
assumption, do you think?
a. Answer: All groups apply equally as fast, because they are motivated on the same
level
6. Draw the results for each of the type of decisions in a figure, with on the x-axis the four
categories (I – IV) of high-school exam results, and on the y-axis the chance of getting
accepted
a. Answer: See chart in type of decisions made
7. Which choice would you make if you were the university? And if you were one of the
pupils? Present your arguments with your answers.
a. University: Optimal merit selection, you get the students with the highest
average. Therefore, the students with the higher success rate as well as the best
performing student. A university wants a lot of students to graduate since they get
a subsidy from the government when their pupils graduate.
b. Pupils: First come first chosen, because then it is not dependent on the grades and
only on being motivated enough to apply first.

Category I II III IV

Average high-school exam X <6 6<X<7 7<X<8 X>8


marks

Number of applications 75 150 100 75

Four types of applications

1. Weighted random selection:

Chances 0.13 0.24 0.45 0.79

Number of admitted pupils 0.13*75 = 9.75 .24*150 = 36 0.45*100 = 45 0.79*75 = 59.25


10 admitted 36 admitted 45 admitted 59 admitted
pupils pupils pupils pupils

2. Optimal merit selection:

Chances 0.00 0.00 0.75 1.00

Number of admitted pupils No pupils No pupils 0.75*100= 75 1*75 pupils


admitted admitted 75 admitted 75 pupils
pupils admitted

3. Random selection:

Chances 150:400 = 0.375 150:400 = 0.375 150:400 = 0.375 150:400 = 0.375

Number of admitted pupils 0.375*75 = 0.375*150 = 0.375*100 = 0.375*75 =


28,125 56.25 37.5 28.125
28 admitted 56 admitted 38 admitted 28 admitted
pupils pupils pupils pupils

4. First come, first chosen:

Chances 150:400 = 0.375 150:400 = 0.375 150:400 = 0.375 150:400 = 0.375

Number of admitted pupils 0.375*75 = 0.375*150 = 0.375*100 = 0.375*75 =


28,125 56.25 37.5 28.125
28 admitted 56 admitted 37 admitted 28 admitted
pupils pupils pupils pupils
Problem 3 - Random selection versus selection with a test
b = (P(VP) = (BR * SR + phi √BR * (1-BR) * SR * (1 - SR)
I Use the computational formula of the above phi coefficient, to calculate the two decision
outcomes under the following circumstances:
Circumstance A:

Success/Satisfactory/Suited A 25 B 475
500 --> .50 P (FN) .025 ( = 25:1000) P(VP) .475 ( = 475:1000)

Failure/unsatisfactory/ C 25 D 475
Unsuited P(VN) .025 ( = 25:1000) P (FP) .475 ( = 475:1000)
500 --> .50

Reject 50 - .05 Accept 950 - .95


b = (P(VP) = (BR * SR + phi √BR * (1-BR) * SR * (1 - SR)

1000 candidates, 500 suited applicants, 950 applicants are accepted


1. Candidates are accepted and rejected on an entirely random basis.
a. You don’t use any test battery → no selection procedure. That is why phi is
zero. The test battery doesn’t take someone’s grades into account and it
doesn’t measure their motivation either.
b. Everything = 0
2. For the selection of candidates, a test battery is used with a predictive validity of .180.
a. BR = a + b = .025 + .475 = .500
b. SR= b + d = .475 + .475 = .950
c. Test validity = .180
b = (.500 * .950) + .180 √.500 * (1-.500) * .950 * (1 - .950) = .478
d. b= .478
e. Proportion valid outcomes: b + c = .475 + .025 = .500
f. Success ratio (SU) = b/ (b + d) = .478/(.478+.475) = .502
3. What is the improvement in the proportion of correct decisions with a predictive validity
of .180 in comparison to completely random selection? And the improvement in SU?
a. First we fill in the formula for a validity of .000 → this is the validity for a
completely random selection. We also calculate the SU as well

b = (.500 * .950) + 0 √.500 * (1-.500) * .950 * (1 - .950) = .475


Proportion valid outcomes: b + c = .475 + .025 = .500
Success ratio (SU) = b / (b + d) = .475 / (.475 + .475) = .500

b. Then we compare this to the formula with a predictive validity of .180


b = (.500 * .950) + .180 √.500 * (1-.500) * .950 * (1 - .950) = .495
Proportion valid outcomes: b + c = .478 + .025 = .540
Success ratio (SU) = b/ (b + d) = .478 / (.478+.475) = .521
c. So, if the predictive validity is .180 then the proportion valid outcomes and the
success ratio (SU) stay the same. Because a predictive validity of .180 is not that
high. If the predictive validity would have been a higher number say .920 then the
equation after Phi would have caused for a higher b. Now this wasn’t seen
because Phi was close to 0.
Increment in percentage correct decisions is 4% [0.04; namely from .50
to .54]; increment in SU = 2% [0.02; namely from .50 to .52]. These values
are low. This has to do with the much too high SR: almost everyone has to be
hired. You can see that the SU under (2) hardly deviates from the base rate
(=% suitable if everyone would have been accepted).

Circumstance B:

Success/Satisfactory/Suited A 38 ( = .04 * 950) B 2 ( = .04 * 50)


40 → (40:1000) = .040 P (FN) .038 ( = 38:1000) P(VP) .002 ( = 2:1000)

Failure/unsatisfactory/ C 912 ( = .960 * 950) D 48 ( = .960 * 50)


Unsuited p(VN) .912 ( = 912:1000) P (FP) .048 ( = 48:1000)
960 → (960:1000) = .960

Reject 950 - .950 Accept 50 - .05

1000 candidates, 40 suited applicants, 50 applicants are accepted


1. Candidates are accepted and rejected on an entirely random basis.
a. You don’t use any test battery → no selection procedure. That is why phi is
zero. The test battery doesn’t take someone’s grades into account and it
doesn’t measure their motivation either.
2. For the selection of candidates, a test battery is used with a predictive validity of .180.
a. BR = a + b = .038 + .002 = .040
b. SR= b + d = .002 + .048 = .050
c. Test validity = .180
b = (P(VP) = (BR * SR + phi √BR * (1-BR) * SR * (1 - SR)
b = (.040 * .050 + .180 √.040 * (1-.040) * .050 * (1 - .050)
d. b= .003
e. Proportion valid outcomes: b + c = .003 + .912 = .915
f. Success ratio (SU) = b/ (b + d) = .003/ (.003 + .048) = .059
3. What is the improvement in the proportion of correct decisions with a predictive validity
of .180 in comparison to completely random selection? And the improvement in SU?
a. First we fill in the formula for a validity of .000 → this is the validity for a
completely random selection. Then b = 0.002, if we calculate b with phi
= .180 you get b = 0.003.
b = (.040 * .050 + 0 √.040 * (1-.040) * .050 * (1 - .050) = .002
Proportion valid outcomes: b + c = .002 + .912 = .914
Success ratio (SU) = b / (b + d) = .002 / (.002 + .048) = .040

b. Then we compare this to the formula with a predictive validity of .180


b = (.040 * .050 + .180 √.040 * (1-.040) * .050 * (1 - .050) =
.010
Proportion valid outcomes: b + c = .003 + .912 = .930
Success ratio (SU) = b/ (b + d) = .003 / (.003 + .048) = .200

c. So, if the predictive validity is .180 then the proportion valid outcomes and the
success ratio (SU) stay the same. Because a predictive validity of .180 is not that
high. If the predictive validity would have been a higher number say .920 then the
equation after Phi would have caused for a higher b. Now this wasn’t seen
because Phi was close to 0.
Increment in percentage correct decisions is 1.6% [0.016]; increment in SU =
16% [0.160]. In short, we are more impressed with the improvements in
terms of SU than in terms of the increment in percentage correct decisions.
SU is more important to companies than b + c, because b + c also includes c,
which refers to correct rejection. Companies are often less interested in c
("we will never see these rejected applicants again"). Because c was already a
good percentage with random selection, the increase in percentage correct
decisions because of using a more valid test battery is less prominent than the
effect in terms of SU.

II Under which of the two circumstances (A or B) is the improvement in the proportion of


correct decisions / SU the highest? Can you explain why this is the case?
Circumstance A
0.6% 0.04 increase for proportion valid outcomes; situation a → (0.503 - 0.500) / 0.500 * 100
= 0.600%

0.4% 0.02 increase for success ratio (SU); situation a → (0.502 - 0.500) / 0.500 * 100 = 0.4%

Therefore, proportion of correct decisions with a validity of .180 - proportion of correct


decisions with a validity of .000) / proportion of correct decisions with a validity of .000 * 100)
● Proportion of correct decisions (validity .180): .503
● Proportion of correct decisions (validity .000): .500
● SU (validity .180) = .502
● SU (validity .000) = .500
Circumstance B
0.1% 0.016 increase for proportion valid outcomes; situation b → (0.915 - 0.914) / 0.914 =
0.100%

47.5% 0.160 increase for success ratio (SU); situation b → (0.059 - 0.040) / 0.040 * 100 =
47.500%

Therefore, proportion of correct decisions with a validity of .180 - proportion of correct


decisions with a validity of .000) / proportion of correct decisions with a validity of .000 * 100)
● Proportion of correct decisions (validity .180): .915
● Proportion of correct decisions (validity .000): .914
● SU (validity .180) = .059
● SU (validity .000) = .040
Circumstance B has a much higher increase in success ratio than circumstances A (47.5% and
0.4% respectively). Circumstances A allows a lot of people who aren’t likely to be successful, in
circumstances B they’re much stricter and the number is lower compared to circumstances A.
The proportion valid outcomes show a small increase in circumstances A than it does in
circumstances B (.1% and .6% respectively), the difference is 0.5%, but this is a small
difference.
Alternative answer to proportion valid outcomes: B was much higher than C in circumstances A
and C was much higher than B in circumstances B, the difference between B and C was greater
in circumstances B. That could explain the difference in increase in proportion valid outcomes.
In Circumstance A, having a test battery improves success ratio (SU) only 2%, while in
Circumstance B, the same test improves SU 16%. This happens due to the changes in base
rate (BR) and selection ratio (SR). That is, when an organization is accepting almost all
applicants and many of them are good (i.e. high SR, high BR – Circumstance A), SU does
not benefit much from test batteries. However, when there is a very strict selection and
only some good candidates (i.e. low SR, low BR – Circumstance B), test batteries could be
very helpful in achieving higher SU. Additionally, the proportion of correct decisions in
Circumstance B has improved only slightly after using the test. This happened since under
random selection so many were already correctly rejected.
Theme 2 - Problems
I Quarrel
There is a fierce discussion among union members and the employer of a large textile company about the
fairness of selection procedures.
Motivate which of these voices would be your voice in this discussion, including why you would not
choose for other voices.
● Voice I: “For me, there is fairness only when all applicants are treated the same. Thus, all
applicants should receive the same tests, the same information about the organization
beforehand, the same interview procedure, etc. I even heard that some applicants are less likely
than others to be shown ads on Google for highly paid jobs."
○ Answer: would not choose, due to indirect discrimination or unintentional. Not everyone
might have the same advantages even when treated the same. Equity is not the same as
equality
● Voice II: “Well, for me fairness will only exist when we consider the applicant’s knowledge,
skills, and experience in our forecasts and not irrelevant information, such as the applicant’s
family connections, values and interests, or even physical attractiveness.”
○ Answer: would not choose, what is relevant information, can depend on the situation and
the job. Once again there could be indirect discrimination (e.g. with GMA). To make sure
discrimination doesn’t happen one should still consider the content of the test and the cut-
off score.
● Voice III: “I disagree. As long as our measures bear a direct or perceived relationship to the job,
our procedure is fair. If selection decisions improve based on the applicant’s values, interests, or
attractiveness these should be taken into account.”
○ Answer: We choose voice III, but we partially agree, participant’s values should be taken
into account, if a participant thinks women should not be doing the job he is applying for
and there are lots of female colleagues, then accepting that person in the job could cause
fights and irritability on the work floor.
● Voice IV: “I surely disagree with all of you. Fairness will only exist when our selection decisions
do not yield decision errors. We should not deny jobs to applicants that could have performed
well.”
○ Answer: would not choose, if you increase the hits and decrease misses, the cut-off
score is higher → the minority group will most likely be less likely to be hired
(Bartam, 1995)
● Voice V: “Well, for me fairness will only exist when our selection decisions do not work to the
disadvantage of minority group members. The selection ratio among the minority group and
majority group should be equal.”
○ Answer: would not choose, because you also have to look at their competence and skills
and when they’re a part of a minority group they might not have the same experience, CV
achievements and connections as to someone who is part of a majority group. So using
the same selection ratios could prove negatively to minority groups. By concealing the
real problems of the imbalances in society it will create inequalities between the groups
(Bartam, 1995).
● Reply by Voice III: “I indeed think we should not base our decisions directly on gender, age or
race. However, I still think that the relationship with the job is a more important factor to
consider. I don’t think that our procedure is unfair when the selection ratios differ, but the
regression lines are equal across both groups.”
○ Answer: would choose (again), this assumes a good selection ratio, because they should
differ.
II Is there Adverse Impact under the below circumstances?
The 80% (four-fifth) rule in which adverse impact conventionally is defined states that:
“A selection ratio for any racial, ethnic, or sex subgroup which is less than 4/5 (or 80%) of the rate for the
group with the highest rate will generally be regarded as evidence of adverse impact”.
A. Given:
a. 722 applicants apply for a job at a governmental service agency
b. 626 applicants are native Dutch (Dutch majority) group members
c. 96 are Dutch ethnic minority group members
d. The governmental service agency accepts 477 Dutch majority group members and 54
Dutch ethnic minority group members.

Native Dutch Dutch ethnic minority Total

Accepted 477 54 531

Rejected 149 42 191

Total 626 96 722

Is there adverse impact against the minority group members under these circumstances?
What is the ratio of the two selection rates? Show your calculations.
Group A (native Dutch): N = 626 → 477/626 = .762
Group B (minority group): N = 96 → 54/96 = .563

Adverse impact = SR (minority group)/SR (native dutch) = .563/.762 = .739


.739 < .80, there is an adverse impact because it is below 80%.

B. As follows:
a. Assume that the highest selection rate in an organization is for the majority group and
that its (the majority's) selection rate is 0.70. Above what value should the selection rate
for the minority group stay to prevent that adverse impact will be claimed?
i. x/.70 = .80/1 → x = .7 * .8 = .56 → this means that the SR should stay above
56% of the applicants.
b. Another organization is more selective in its hiring: the majority selection rate is 0.20.
Below which selection rate for the minority group will adverse impact be claimed?
i. X / .20 = .80 → x = .20 * .80 = .16 → this means that the SR should stay above
16% of the applicants.
c. Which of the two organizations has more room for the two selection ratios to ‘drift’ apart
before adverse impact is claimed? Explain why this is the case.
i. Group B, 1- .16 = .84. Therefore, it has a higher number to drift apart compared
to (1 - .56 = ) .44

III How to use Bartram's tables on Adverse Impact


Use Tables 1 and 2 of Bartram (1995), see Appendix II.
1. Given in a group of 200 applicants, there are 196 majority and 4 minority applicants. The
majority applicants on average score 1/5 standard deviation higher than the minority applicants.
If the organization selects the 50 highest scoring applicants, will the organization meet the 80%-
rule? Explain why it will (not). Show your calculations.
a. Answer: 4/196 = .02 is the ratio of minority applicants to majority applicants.
And we know the SD = .2 (⅕). To select the table 196/200 = .98 → table 1B.
If you then look in the table for 98% of the applicants in the higher group, and the z-
difference score of .2, you find a selection ratio of 35.58 → 36%. (.36*200) = 72
→ %sel_B/ %sel_A = 28.57/35.72 = .799 → just below the 80% rule!

2. An organization has 125 applicants from an ethnic minority group who on average have a
lower test score than the 375 applicants from the majority group. The organization can
maximally select 50 applicants as it has 50 vacancies. Show, given this number of vacancies,
under which circumstances the organization will and under which circumstances the
organization will not get into trouble with the 80%-rule.
a. Answer: 125+375=500 is the total amount of applicants with 50 possible selections. The
ratio of accepted candidates is 50/500 = .1, the percentage of applicants in the higher
scoring group are: 375/500 = .75 (table 2e). You need a z-diff of 0.0 or 0.1 to not have an
adverse impact ratio (above 80% rule).
b. If the z-diff is above 0.1 then there is an adverse impact ratio (under 80% rule). So,
according to table 1e (%selected): 0.0345 * 50 = 1.725; Therefore you need to accept at
least 2 people from the minority group to not have an adverse impact ratio.
3. An organization has 30 applicants from an ethnic minority group who on average have a lower
test score than the 270 applicants from the majority group. The organization can maximally
select 120 applicants. Show, given this number of vacancies, under which circumstances the
organization will and under which circumstances the organization will not get into trouble with
the 80%-rule.
a. Answer: 270+30=300 is the sum of applicants from minority and majority groups with
120 possible selections. The ratio of accepted candidates is 120/300 = .4, and we know
what the ratio of the majority group which is: 270/300 = .90, that’s why we use table 2D
(90% of applicants in the higher scoring group) is the ratio of minority applicants to
majority applicants.
You need a z-diff of 0.000, 0.100 or .200 to not have an adverse impact ratio (above 80%
rule). If the z-diff is above 0.2 then there is an adverse impact ratio (under 80% rule).
b. If the z-diff is above 0.2 then there is an adverse impact ratio (under 80% rule). So,
according to table 1e (%selected): 0.3486 * 120 = 41.832; Therefore you need to accept
at least 42 people from the minority group to not have an adverse impact ratio.
4. Which values will the selection ratio generally have if the difference between two groups (say a
majority and a minority group) equals or is larger than 0.5 standard deviation and at the same
time the organization meets the 80%-rule? What do you conclude from this?
a. Answer: if we look at a z-diff (SD) of .5 then we can conclude a SR needs to be above 84
(rounded up) to meet the 80%-rule (if we’re looking at table 2a). If the SD gets higher,
the SR will also get higher, but the impact is lower.
Theme 3 - Problems
Objective
Ultimately, the client-organization will want to know whether the selection procedure used really adds
value. That is, the client-organization wants to know whether the selection methods used result in better
employee productivity and whether they are worth the effort? This issue is labeled utility analysis. The
theme talks about three well-known models to estimate utility.

Problem 1 - The Taylor-Russell utility model


Make use of the Taylor-Russell Tables (Appendix III) to answer the following questions:
1. The tables present the proportion suitable applicants after selection (the success ratio SU) in the
heart of each table as a function of:
a. the proportion suitable applicants without (or: before) selection (the base rate BR)
b. the selection ratio SR
c. the correlation between predictor and criterion (rxy)
d. Fill out the following Table 1

Validity rxy SR (selection rate) BR (base rate) Success Ratio SU

0.25 0.20 0.30 .43

0.55 0.70 0.80 .89

0.20 0.70 0.80 .83

0.10 0.50 0.50 .53

0.55 0.50 0.50 .69

2. For the following four questions, use the data from the example below in Table 2, unless
specified otherwise.
Table 2. Example.
Interpretation criterion:
1 = insufficient productivity
2 = doubtful productivity
3 = sufficient productivity
Of the 20 individuals, 10 have been selected.
Questions:
1. If individuals with a criterion score equal to '3' were rated as 'suitable',
what would the proportion suitable applicants before selection (BR) be?
Assume that rxy = .70.
a. Answer: the selection ratio is 10/20 = .50 and the correlation is rxy = .70 (given).
b. In the table we see that 5 people score (individuals: 13, 16, 17, 18, & 20) a criterion score
equal to ‘3’. The success rate = 5/10 = .50
Success ratio (SU) = b/ (b + d) = 5 sufficient productivity/ (5 sufficient productivity + 5
who score lower) = .50
c. We look at the table in the row of .70 and .50. The table which is closest to .50 (success
ratio) shows that the base rate is .30, because in this table the combination of the rxy and
the selection ratio is .51 (which is closest to a success ratio of .50)
2. Give the answer to question 1, what is the 'gain' in terms of SU as a result of selection? Compare
the situation in which selection was used, with the situation in which applicants are admitted
randomly.
a. Answer: compare rxy = .00 with rxy = .70.
b. When rxy = .00 (random selection) then the success ratio is .30 (equal to the base
rate). We can say now that there is an increase of .51 - .30 = .21 → 21 percentage
point increase.
3. What is the proportion of suitable people after selection in the group that is rejected (that is to
say, not selected)?
a. Answer: 20 * 0.3 (base rate) = 6 people are suitable in the total sample but we know that
only 5 people were suitable in the hired sample.
b. 6 people suitable individuals in total sample - 5 selected suitable individuals = 1 → so
this one person is ‘left’ in the rejected group (which is suitable).
c. 1 / 10 = .10, so 10% of the people were rejected but also suitable
4. If individuals with a criterion score is equal to '2' also are rated as suitable, how many 'suitable'
individuals would one have rejected in this selection procedure?
a. Answer: the selection ratio is 10/20 = .50 and the correlation is rxy = .70 (given).
Individuals with a score equal to 2 are 4, and with a score equal to 3 are 5, 4 + 5 = 9
individuals who are suitable.

b. Success ratio (SU) = b/ (b + d) → 9/10 = .90

c. Look in the tables → which is closest to .90 (success ratio) shows that the base rate
is .70, because in this table the combination of the rxy and the selection ratio is .91
(which is closest to a success ratio of .90)

d. So the base rate of .70 would lead to .70 * 20 = 14 people who are suitable in the total
sample. We know that 9 people in the selected sample are suitable → 14 - 9 = 5
people in the rejected sample (who would have been suitable for the job). This gives
a percentage of 5/10 = .50 → 50% is wrongly rejected.

Conclusion: The stricter you are the less people will be wrongfully rejected (higher criterion score).

Problem 2 - The Brogden-Cronbach-Gleser utility model


1. Use the Brogden-Cronbach-Gleser (BCG) model. Find the net gain in utility (ΔU ) over random
selection a) overall and b) per selectee, given the following information:
Answer:
(125 * .35 * 7500€ * .3988) / .5 - (125 * 400€) / .50 = 261712.50
- 25000 = 236712.50
So, the overall utility is 236,712.50.
Per selectee it is: 236,712.50 / 125 = 1,893.70

2. Given the following information on two selection procedures, and using the Brogden-Cronbach-
Gleser model, what is the relative difference in payoff a) overall and b) per selectee between the
two procedures? For both procedures, quota Ns = 50, SR = 0.50, λi = 0.3939, and Sdy = Euro
15000,=.

Circumstance 1:

Ry1: 0.20; C1 = Euro 100,= → less valid and less expensive

Overall: ((50 * .20 * 15000 * .3939) / .50) - ((50 * 100) / .50) = 108,170
Per selectee: 108170. / 50 = 2163,40

Circumstance 2:

Ry2: 0.40; C2 = Euro 500,= → more valid and more expensive

Overall: ((50 * .40 * 15000 * .3939) / .50) - ((50 * 500) / .50) = 236340 - 5000 = 231,340
Per selectee: 231,340 / 50 = 4,626.80

Difference:

Overall: 108,170 / 231,340 = .468 → 46,8%, circumstance 2 delivers 46.8% of the utility that you get
with circumstance 1

Selectee: 2163,40 / 4,626.80 = .468 → 46,8%, circumstance 2 delivers 46.8% of the utility that you get
with circumstance 1

It doesn’t matter how costly the test is, in circumstance 2 it doesn’t cost as much. The test is more valid in
circumstance 2 (and also more expensive). The expenses are not as important as the validity of a test, so
it’s preferable to have a high validity as compared to lower expenses. Nowadays companies first look at
the expenses and not so much at the validity, but the people you’re selecting will be of less value to the
firm then.
3. Provide a worked example using the BCG-utility formula in which you create and compare two
scenarios:
a. Show how a selection test with a given non-zero predictive validity and a given Ns will
result in a utility which differs between the two scenarios because the scenarios have
different values on one or more other components of the BCG-formula.

Circumstance 1 - Ry2: 0.70; C = Euro 300, Ns = 50, SR = 0.50, λi = 0.3939, and Sdy = Euro 80,000=
Overall: ((50 * .70 * 80,000 * .3939) / .50) - ((50 * 300 / .50)= 2,205,840 - 30,000 = 2,175,840
Per selectee: 2,175,840 / 50 = 43,516.80

Circumstance 2 - Ry2: 0.70; C = Euro 10, Ns = 50, SR = 0.50, λi = 0.3939, and Sdy = Euro 10,000,=
Overall: ((50 * .70 * 10000 * .3939) / .50) - ((50 * 10) / .50) = 236340 - 5000 = 274,730
Per selectee: 274,730 / 50 = 5,494.60
Differences:

Overall: 274,730 / 2,175,840 = .126 → circumstance 2 delivers 12.6% of the utility that you get with
circumstance 1

((New - old) / old) * 100 = percentual increase = ((274,730 - 2,175,840) * 2,175,840) * 100 = -87.374%
→ It’s a decrease of 87,374% if we use the measure from circumstance 2

Per selectee: 5,494.60 / 43,516.80 = .126 → 12.6%

b. Explain your findings, including an explanation of the most favorable scenario for the
selection test to be useful.
Answer:
- Costs of the predictor and its administration: utility will be higher when the costs are lower.
But when you have an high income as an organization, it is ok to pay more for a selection test
because you can afford it as a firm (e.g. a function at google vs a function at your local bookshop)
- Validity coefficient: selection procedures with a high validity because then the utility will also
be greater.
Problem 3 - Use the Taylor-Russell utility tables to save the police money
A police force wants to recruit 100 constables. It is a difficult job and therefore only 30% of the
workers would perform satisfactory if they were selected from applicants at random. There are 500
applicants. A combination of situational interviews and written tests with a validity of 0.60 is used.
A commissioner in charge of the finances calculates that each recruit who subsequently fails training
costs Euro 10.000,
Given this information:
How many Euro's will the force's selection system of situational interviews and tests save for 100 people
it recruits in comparison to a random selection procedure?
Answer:
Base rate = .30; Validity of the test = .60; Selection Ratio = 100 / 500 = .20
Success Ratio = .64 (look at the Taylor-Russell table: row .6 and the column of .2 in the table with a base
rate of .3)
.64 * 100 = 64 individuals, random selection .30 (base rate) * 100 = 30 individuals
Using the measure you’ve 34 (64 - 30) people more fit for the job compared to random selection.
So that’s 34 people less to fail the training, you save 34*10 000 = 340.000 euros compared to using
random selection.
So, using a measure costs less than failing a test. It’d be better to use a measure in this case, since the test
has a high validity (.60) and using random selection you only have a success ratio of 30%.

You might also like