You are on page 1of 3

Translated from French to English - www.onlinedoctranslator.

com

3. A practical way to present a correlation: the table of expectations.


Thecorrelationis aindicatorstatistic which is notnotstilleasily understoodby non-
statisticians. In some situations, it is important toto presentthemresultsof a correlation
analysis in a moresimple. The table of expectations is the ideal tool to do this.
Example :
Consider the following situation: In a large call center, employees' work performance is
determined by the number of customers they serve in a day. The center wants to improve
the system it uses to select future employees, that is, to choose those who will be able to
serve more customers.
The vice-president of the center asks a researcher to develop a new system for selecting
candidates, which he does by developing a test tomeasure work ability. The measure of
ability is the variableX. Theperformancetowork is the variableY. The researcher
hypothesizes that the people who obtain the highest values on the test will prove to be the
most productive at work. He intends to verify it by calculating the correlation between the
measurement of aptitude (X) and performance at work (Y).

To verify his hypothesis, he randomly chooses 180 people already in office. Company
records provide her with their job performance: the average number of customers each has
served, each day, over the past month. This measurement varies between 20 and 80. He
administers the work aptitude test to these 180 employees. He therefore obtains, for each
person, two pieces of information: his performance at work (Y) and his performance on the
aptitude test (X). He verifies the hypothesis by calculating the XY correlation and finds that it
is positive and substantial: rxy= 0.58. The researcher now has proof that performance on the
aptitude test is positively related to job performance. Thus, those who demonstrate the
greatest ability (as measured by the test) tend to be more productive. It can also be said that
knowing the ability to work (X) reduces uncertainty about the eventual performance at work
(Y).
Technically, the researcher performed a concurrent validity study. For this kind of study, a
correlation of 0.58 is considered very substantial and psychometricians would say that the
test is a "valid" measure of job performance.

He must now communicate the result of his study to the vice-president of the centre. She
is not a statistician and a correlation of 0.58 will mean very little to her. The researcher then
chooses to present him with the XY correlation he has obtained in a table of expectations.

Atable of expectationsis amatrixatdouble entrythat the researcher constructs as


follows: he divides the employees who participated in his study into three groups of 60
people each. In thegroup 1, which he labels "Low performance»,it places people in the lower
third of the job performance distribution. He places in thegroup 3, “High performance»,
people who are in the top third of job performance. All the others, the third of his sample
that falls in the middle of the job performance distribution, are placed in the group 2, “
Average performance".

The scores obtained on the aptitude test vary between 20 and 80. The researcher then
divides the performance on the test into three groups: thegroup 1, “Low ability»,includes
employees who scored 39 or less on the test. Those who roll 60 or more form the group 3: “
High aptitude".The others, those who scored between 40 and 59, form the group 2: “
Average ability".Table 7 shows the observed data. Note that, in this table, 60 employees are
classified in the “Low ability” group, and 56 and
2

64 employees, respectively, in the “Medium ability” and “High ability” groups.

Next, the researcher identifies for each aptitude test performance group the number of
people who have low, average or high job performance. We note in Table 7 that, of the 60
people who scored low on the aptitude test, 45, 14, and 1 fall into the low, medium, and
high performance groups, respectively (row 1 of the table). We can now express these
results in percentages (shown in probabilistic form in parentheses in the table). Thus, we see
that 75% of people who scored low on the test show low job performance, and only 2% of
people who show low ability show high job performance. About a quarter (23%) of people
who scored low on the test provide average job performance. Interpreting these
percentages in probabilistic terms,we can conclude that those who scored low on the test
have a very low probability (p = 0.02) of providing strong job performance, an intermediate
probability (p = 0.23) of being moderately productive, and a very high probability of
performing poorly at work (p = 0.75)

Table 7: The expectations table

Work performance (Y)

Ability (X) Grp 1 (Low) Grp 2 (Medium) Group 3 Total


(High)

Weak 45 (0.75) 14 (0.23) 1 (0.02) 60

Medium 13 (0.23) 29 (0.52) 14 (0.25) 56

Raised 2 (0.03) 17 (0.27) 45 (0.70) 64

TOTAL 60 60 60 180

We perform these analyzes for each row of the expectations table. Take the third
row of data in Table 7 for example: of the 64 people who demonstrated strong
performance on the test, 3% (2/64) are low productive, 27% (17/64) are moderately
productive and 70% (45 /64) are very productive. Expressing these percentages in
probabilistic terms, we can conclude that people who do very well on the test (high
aptitude; 60 and above) have a very high probability (p = 0.70) of being very productive
employees (group 3 ) and a very low probability (p = 0.03) of performing poorly at work.

If the vice-president decides to administer this aptitude test to applicants, we can see, by
consulting the table of expectations, that it would be preferable not to hire the candidate
who obtains a low score (< 40) on the test. because he will have a low probability of
providing an exceptional work performance (p = 0.02) and a very high probability (p = 0.75)
of not performing well. But if he obtained more than 59, his probability of becoming a very
productive employee would be very strong (p = 70) and it would then be relevant to hire
him.
3

In fact, thepictureof theexpectationsonly serves {reproduce, in terms that it is moreeasyof to


understandand put inpractice,informationalreadyestablishedover therecorrelation:bigger is
aptitudeof one person, morehighwill be hisperformanceat work.
A question might now come to mind: if the expectation chart is a convenient and simple
way to show the correlation between two variables, why did we calculate the correlation
(and have you study an entire chapter on the subject)? The answer brings us back to the
discussion in Chapter 1 on measurement scales.The measure ofaptitudeand that of the
performanceat work in Table 7 are scales tointervals. The table ofexpectations translated
these variables into scalescategorical(nominal).
As we saw in Chapter 1, theconversionof a scale {intervalsin a ladder nominal reduced
theprecisionDatas. Thus, the categoryLow ability»encompasses both the person who
scored 20 on the test and the one who scored 39, and considers the latter to have
performed very differently from another person who would have scored 40, only one point
higher. Thus, thecategorization causesalossimportant information. In the next chapter, we
will study another technique, thesimple regression, which allows us to do the same kind of
prediction without converting the data to nominal variables. But, for that, you will have
to learn and understand other statistical techniques! Summary of Chapter 6

The Pearson correlation is aindexstatistic that indicatesthe degree of similarity between the
position of the observations on one variable and the position of these same observations on a
second variable.
Shelimitto indicate thedegreeoflinear relationshipthat exists betweentwo variableswhich
are measuredwith somescalesatintervalsor scales ofreport. The correlation takes values
ranging from 0 to ±1.0. The relationship isperfectwhen it is equal to ±1.0 and itreducedas it
gets closer to0.
thesignof the correlation indicates thedirectionof the correlation. Correlation and its
coefficients ofdeterminationand ofNope-determinationare used to interpretto what extent
theknowledgeof onevariablereduces uncertainty in the face ofsecond variable. The correlation
ofPearsonisinfluencedby the values beingfarof themean, and is relatively unaffected by
commentstolocated nearfrom her. Finally, thepresenceof onecorrelation does not indicatenot
necessarilythepresencefrom a linkcausalbetween variables. Butthe absence ofcorrelation
indicates aabsenceofcausality.
Thecorrelationis part ofdescriptive statisticswidely used in sciencehumanand in science
social, mainly because itnotedto what extent theknowledge of onevariableweinformabout
asecond variable. Finally, it is possible to present the correlation between two variables in a
formmoresimple, the table ofexpectations.

CORRELATION Applied Statistics 21/11/2022

You might also like