HR Analytics
Session 10
Anshul M
GIM
Recap
• Missing value analysis
• Listwise vs pairwise deletion
• Single imputation vs multiple
• T-test
Decision-Making Process
Absolute zero
Skewness measures the degree and Kurtosis measures the “peakedness” or tail
direction of asymmetry in a data heaviness of a distribution compared with a norm
distribution relative to the mean. distribution.
•Skew: To calculate skewness, use the •Kurt: To calculate kurtosis, use the
function skew(data set). function kurt(data set).
•If the skewness is between -0.5 and 0.5, the •In a normal distribution, the kurtosis
data are fairly symmetrical. If the skewness is is equal to three. Kurt>3, higher peak and
between -1 and – 0.5 or between 0.5 and 1, positive. Kurt<3, flatter peak and
the data are moderately skewed. If the negative
skewness is less than -1 or greater than 1, the
Application for HR
• Salary Distribution Analysis Performance Appraisal Consistency
If performance ratings display high
• Positive skew (long right tail) suggests kurtosis, it indicates that most ratings
that a few employees earn significantly
cluster near the mean with very few
higher pay than the majority, which
may indicate pay inequality. extremes. This may signal:
• Managers avoiding differentiation among
• Negative skew (long left tail) might employees (central tendency bias).
reflect pay compression or capped • Limited use of the full rating scale in
salary ranges.
evaluations.
• Performance Ratings
• A highly positive skew in ratings might Salary and Compensation Review
suggest rating leniency, while a • High kurtosis means salaries are clustered
negative skew could signal managerial near the average, possibly showing pay
strictness. Both patterns affect compression.
promotion and bonus fairness. • Low kurtosis indicates large variation in
salaries, which may suggest inequity or a
Correlation
• Correlation is a statistical measure that expresses the
extent to which two variables are linearly related.
• A sample correlation coefficient is called r, while a
population correlation coefficient is called rho, the
Greek letter ρ.
=CORREL
Spurious correlation
• In statistics, a spurious relationship or spurious correlation is
a mathematical relationship in which two or more events or In East Asian corporations,
variables are associated but not causally related, due to many buildings avoid the
either coincidence or the presence of a certain third, unseen fourth floor because the
factor. number four sounds similar to
• Superstition the word for death.
• Superstition refers to the attribution of causal power to
events or actions that have no logical or empirical basis. It At several technology firms,
stems from psychological or cultural beliefs rather than teams keep “lucky objects”
statistical relationships.
such as toys or figurines
• An employee believes that wearing a specific color on
near servers or
appraisal day brings better evaluation results.
workstations.
• These behavior arises from the human tendency to find
patterns in random events, particularly under uncertainty.
• In organizations, such beliefs may influence decision-making,
Machine operators in some
rituals, and perceived control, even when unsupported by factories perform personal
evidence rituals before beginning a
shift, believing it will prevent
breakdowns.
Spurious correlation
Guess
• Number of accidents and number of parties in Goa in
winters
• Money in office renovation and reduced conflict with
manager
• Breakfast time in morning and academic performance
Measure of Data
• Self-report measures: Usually questionnaires or
interviews that measure how people report that they
act, think, or feel
• Behavioral- How often you do something?
• Cognitive self-report measures- what they think about
something?
• Affective self-report measures
• Tests: personality tests and ability tests
• Behavioral measures: Measures taken by carefully
observing and recording behavior.- filling it for a
supervisor, subordinate, peer
• Physical measures: Measures of bodily activity (such as
Reliability
• Reliability refers to the consistency or stability of a
measuring instrument. In other words, we want a
measure to measure exactly the same way each time it
is used. This means that individuals should receive a
similar score each time they use the measuring
instrument.
• Reliability is measured using correlation coefficients.
Validity
• Validity refers to whether a measure is truthful or
genuine. In other words, a measure that is valid
measures what it claims to measure.
• Validity refers to the degree to which a test or
measurement accurately measures what it is intended
to measure.
• Example:
• A valid driving test should include a practical driving
component and not just a theoretical test of the rules of
driving.
Types of Validity
[Link] Validity: Reflects the complete content of
the construct.
[Link] Validity: Measures the theoretical
construct accurately.
[Link]-Related Validity: Relates to predicting
outcomes based on other measures.
[Link] Validity: Focuses on future performance.
[Link] Validity: Compares results with another
established measure.
[Link] Validity: Appears relevant and appropriate at
face value.
Type of
Definition Example
Validity
The extent to which a measurement reflects the A math test covering all
Content
entire content or domain of the construct being relevant topics in the
Validity
measured. curriculum.
Constru A personality test
The degree to which a test measures the
ct accurately reflecting traits
theoretical construct it claims to measure.
Validity like extraversion.
Criterio
A job performance test
n- The effectiveness of a measure in predicting
correlating with employee
Related outcomes based on another measure.
success rates.
Validity
Predicti
The extent to which a test predicts future SAT scores predicting
ve
performance or outcomes. college success.
Validity
Concurr The degree to which test results correlate with A new depression scale
ent results from a recognized measure taken at the showing similar results to a
Validity same time. standard scale.
The extent to which a test appears to measure A survey on eating habits
Face
what it is supposed to measure, based on looks reasonable to
Validity
subjective judgment. participants.
Reliability and Validity for HR
Managers
Qualitative Interview design
1. Research Design and
Approval
2. Interview Opening
Establish rapport
Confirm consent
Context setting
3. Core Interview Process
Question sequencing:
background → experiences
→ reflections → conclusions
Probing and clarifying
Neutrality maintenance
Recording and note-taking
4. Closing the Interview
Summary and reflection
Appreciation and debrief
Request for further
clarification
Post-interview notes
5. Interview Closure and
Research Exit
HR Assessment after Leadership
Transition at X
• When Elon Musk completed his acquisition of Twitter for 44 billion USD in October 2022, he
moved quickly to remake the company’s structure, culture, and strategy. Within days, he
dismissed top executives including CEO Parag Agrawal and dissolved the board of directors.
The company shifted from a public corporation to a privately held entity under Musk’s
control.
• A series of rapid operational changes followed. Nearly half the workforce was laid off in
November 2022, followed by additional reductions that left the company with a fraction of its
original staff. The moderation, policy, and communications teams were significantly reduced.
Musk justified these decisions as necessary cost-cutting to stabilize finances and accelerate
innovation.
• Platform governance changed markedly. The verification system was replaced by a paid
subscription model, moderation rules were relaxed, and advertising partners expressed
concern about brand safety. Many major advertisers paused spending, causing a sharp
decline in ad revenue.
• In 2023, Musk rebranded Twitter as X, positioning it as the foundation for an “everything
app” integrating social networking, payments, and commerce. The rebranding reflected his
ambition to create a multifunctional digital ecosystem rather than a conventional social
media platform.
• The immediate challenge is to assess the organizational climate, leadership
effectiveness, and employee resilience following these changes.
Sampling
1. Probability Sampling them?
• Simple Random Sampling: Every
member of the population has an equal
2. Non-Probability Sampling
chance of being selected.
• Convenience Sampling: Selecting
• Systematic Sampling: Selecting every
individuals who are easiest to reach or
nth member from a list of the population.
sample.
• Stratified Sampling: Dividing the
• Judgmental (Purposive) Sampling:
population into subgroups (strata) and
Selecting individuals based on the
taking a random sample from each.
researcher’s judgment about who would be
• Cluster Sampling: Dividing the population most informative.
into clusters (often geographically) and
• Snowball Sampling: Existing study
randomly selecting entire clusters for study.
subjects recruit future subjects from among
If we are doing a study on students academic their acquaintances, useful for hard-to-
experience after term 1 at GIM, should we reach populations.
have a simple random sampling or something
• Quota Sampling: Ensuring equal
else?
representation of specific characteristics by
How should we design an experiment for setting quotas for different subgroups.
Significance level
• The significance level (α) in statistics is the
probability of rejecting the null hypothesis (H₀) when it
is actually true. In simpler terms, it represents the
likelihood of making a Type I error, which occurs when
you conclude that there is an effect or difference when
in reality there isn’t.
• α = 0.05: This means you're willing to accept a 5%
chance of incorrectly rejecting the null hypothesis.
• If the p-value (calculated probability) of a test is less
than or equal to the significance level (e.g., p ≤ 0.05),
you reject the null hypothesis.
• If p > α, you fail to reject the null hypothesis.
•In medical trials or critical applications, a lower α (e.g., 0.01) is often chosen to minimize
the risk of false positives.
•In exploratory research, a higher α (e.g., 0.10) might be acceptable to avoid missing
potential discoveries (false negatives).
Power in Stats
• In statistics, power refers to the probability of correctly rejecting the
null hypothesis (H₀) when it is actually false, or in other words, the
ability of a test to detect an effect when there is one. Power is
influenced by factors such as sample size, significance level (α), effect
size, and population variability.
• Power is typically denoted as 1−β1, where:
• β (Beta) is the probability of making a Type II error (failing to reject the null
hypothesis when it is false).
• Power = 1 - β.
• The recommended value for statistical power is typically 0.80, or
80%. This means there is an 80% probability of correctly rejecting the
null hypothesis when it is false (i.e., detecting a true effect). The risk of
making a Type II error (failing to detect an effect when one actually
exists) is 20% (β = 0.20)
Factors Influencing Power:
1. Significance Level (α): The probability of making a Type I
error (rejecting the null hypothesis when it is true), commonly
set at 0.05.
2. Sample Size (n): Larger samples provide more information
and increase the power.
3. Effect Size: The magnitude of the difference between the null
hypothesis and the true population parameter. A larger effect
size increases the power.
4. Population Variability: Less variability (smaller standard
deviation) increases the power of a test.
[Link]
Standard Error and confidence
Interval
x̄ represents the sample mean
z* is the critical value from the
standard normal distribution
Sampling bias
•Undercoverage/selection bias- The bias comes from missing groups
• Sampling bias before sampling begins.
occurs when An HR department studies employee engagement but only surveys office-
certain members or based employees, excluding remote and field staff
groups in a •Self-Selection bias- The bias arises because participation is a choice
population are correlated with attitudes or ability.
more likely to be A diversity perception survey has higher representation of female workforce.
included in a •Nonresponse Bias- The bias comes from nonparticipation among
sample than
those already selected.
others, leading to a
sample that is not Exit survey responses come mostly from mostly dissatisfied employees
representative of •Healthy Worker Effect- The bias is in comparing an employed sample
the entire to an inappropriate external population.
population. A work stress survey is only field by employees who didn’s use medical claim
in last few years or had low sick leaves.
• This can result in
•Convenience Sampling Bias- The bias stems from choosing what is
inaccurate or
easy to reach, not from nonresponse or voluntary participation.
misleading
HR collects satisfaction data from those attending in-person town halls
conclusions.
Non-response
Bias
Rogelberg, S. G., & Stanton, J. M. (2007).
Introduction: Understanding and dealing with
organizational survey nonresponse. Organizational
research methods, 10(2), 195-209.
Survey Research Methods
• A survey is a means of "gathering information Plagued by five important
about the characteristics, actions, or opinions weaknesses: (1) single-method
of a large group of people, referred to as a designs where multiple methods
population are needed
(2) unsystematic and often
• While surveys conducted for research
inadequate sampling
purposes have three distinct characteristics
procedures,
• is to produce quantitative descriptions of some (3) low response rate
aspects of the studied population
(4) weak linkages between units
• by asking people structured and predefined of analysis and respondents,
questions (5) overreliance on cross-
• Information is collected from a fraction of sectional surveys where
population but applicable to whole. longitudinal
Pinsonneault, surveys
A., & Kraemer, K. (1993).are needed
Survey
research methodology in management information
• Classified in three different types- systems: an assessment. Journal of management
information systems, 10(2), 75-105.
• Exploration- become familiar with new topics
• Description- find out situations, events,
attitudes, or opinions
• Explanation- to test theory and casual
Measurement error refers
to inaccuracies or
discrepancies that arise
during the process of
collecting data, often caused
by factors such as biased
questions, poor interviewing
practices, or the withholding
of accurate information by
the interviewees.
• These errors can occur
when participants
misinterpret or deviate
from the intended order
of questions, leading to
different responses.
Recap
Type I and Type II errors
Significance level
– α- rejecting the null hypothesis H0when it is actually true
Power
− β- probability of correctly rejecting the null hypothesis when the
alternative hypothesis H1 is true
Standard error
Confidence interval
Sampling bias
Measurement error
Common method variance
Common Method Variance
• Broadly defined, method variance refers to any systematic, non-substantive influence
on measures of substantive constructs that is due to the method of measurement
used.
• The term method refers to the form of measurement at different levels of
abstraction, such as the content of specific items, scale type, response format,
and the general context (Fiske, 1982, pp. 81–84). At a more abstract level, method
effects might be interpreted in terms of response biases such as halo effects, social
desirability, acquiescence, leniency effects, or yea- and nay-saying.
• Three factors
(a) characteristics of the respondent
(b) properties of the items used to assess a construct
(c) general features of the questionnaire and the questionnaire context influencing
the measurements
Podsakoff, P. M., MacKenzie, S. B., Lee, J. Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: a
critical review of the literature and recommended remedies. Journal of applied psychology, 88(5), 879.
T-test
A t-test is a statistical test used to compare the means
of two groups and determine whether the difference
between them is statistically significant. Types:
• One-sample t-test: When you want to compare a sample
mean to a known or hypothesized population mean.
• Independent t-test: When you are comparing the means
of two independent groups.
• Paired t-test: When you have paired data (e.g., the
same subjects measured twice, or matched pairs).
T-test Assumptions
• 1. Independence: The observations in one sample are independent of
the observations in the other sample.
• 2. Normality: Both samples are approximately normally distributed.
• 3. Homogeneity of Variances: Both samples have approximately
the same variance. (If the ratio of the larger variance to the smaller
variance is less than 4, then we can assume the variances are
approximately equal and use the two sample t-test.)
• 4. Random Sampling: Both samples were obtained using a random
sampling method.
• If one or more of these assumptions are violated, then the results of the
two sample t-test may be unreliable or even misleading.
How to test assumptions and what
next?
• Assumption 1: Independence- Data is correctly collected
• To do: Recollect data
• Assumption 2: Normality
• Check= n < 50, Shapiro-Wilk test, P < 0.05
• N > 50, Q-Q plot, the data should be along the diagonal line
• To do: Mann-Whitney U test, which is non-parametric equivalent and
doesn’t consider normal distribution
• Assumption 3: Homogeneity of Variances
• To do: Welch’s t-test, which is a non-parametric version of the two sample
t-test and doesn’t consider equal variances
• Assumption 4: Random Sampling
• To do: Recollect or don’t generalize the results
Application of t-test in HR
Find the correct one
Research Hypothesis –
A1) H0-It is hypothesized that application of different e-HRM tools
are same to the mean (test) value in Indian organizations.
A2) H0- It is hypothesized that that the application of e-HRM tools
are at the same level for public and private organizations.
A3) H0-- It is hypothesized that that the application of e-HRM
tools are at the same level for manufacturing/ mining and
services organizations
A4) H0-- It is hypothesized that the application of e-HRM tools are
at the same level for all the select Indian organizations.
Interpretation- JS across Mumbai
and Delhi
• t-Statistic (t Stat): The t-statistic is -1.33, indicating the observed difference
in job satisfaction means between Delhi and Mumbai is not very large
• Degrees of freedom: 255 (DF) in statistics indicate the number of
independent values that can vary in an analysis without breaking any
constraints.
• P-Value (One-Tail): The one-tail p-value is 0.093, which is greater than the
typical significance level of 0.05. This suggests that the difference in job
satisfaction between the two cities is not statistically significant at a 5%
level in a one-tailed test.
• t Critical (One-Tail): The critical t-value for a one-tailed test at the 5%
significance level is 1.65. Since the t-statistic (-1.33) does not exceed this
critical value, we fail to reject the null hypothesis in the one-tail test.
• P-Value (Two-Tail): The two-tail p-value is 0.186, also greater than 0.05
• t Critical (Two-Tail): ?
A/B Testing Steps in A/B Testing
[Link] Objective
• A/B testing is a controlled experimental
1. Clearly specify the key outcome (e.g., click-through
approach used to compare two variants (A and B)
to determine which performs better on a defined rate, engagement score, conversion rate,
outcome. It is widely applied in marketing, HR, retention).
product design, and organizational decision-making. [Link] Population
1. Randomly assign participants into two groups to
• Concept
reduce bias.
• Variant A: Often the current or control version. 2. Example: 50% of employees receive onboarding A,
• Variant B: The new or experimental version. 50% receive onboarding B.
[Link] Variants
• Objective: Measure differences in outcomes to 1. Deliver version A to the control group and version
identify which variant is more effective. B to the experimental group.
• Example: Testing two versions of an employee [Link] Data
onboarding program to see which improves early 1. Measure the predefined outcome during the
productivity more. testing period.
Applications in HR and Management 2. Ensure data quality and consistency.
Recruitment: Testing two interview scripts to see which predicts [Link] Results
better job performance. 1. Use statistical tests (t-test, chi-square, or
Employee Engagement: Comparing two communication regression) to determine whether differences are
strategies for internal campaigns. significant.
Training Programs: Evaluating which version of a training 2. Calculate effect size and confidence intervals to
module improves skill acquisition. understand practical relevance.
Performance Incentives: Testing different reward schemes to [Link] Making
see which increases productivity or retention. 1. Choose the variant with superior performance.
[Link]
ab-testing-can-help-transform-the-effectiveness-of-employee-training/
2. Consider replicating the test to validate results
Experimental methods
Element Description in HR Context
• Experimental methods
The researcher intentionally
refer to research designs changes a variable (for example,
in which the investigator Manipulation
feedback type, incentive system, or
manipulates one or more training format).
independent variables to Other potential influences are held
observe their causal Control constant or randomized across
conditions.
effect on dependent
variables, while Participants (employees, managers,
Random or teams) are randomly assigned to
controlling for Assignment different conditions to equalize
extraneous factors. The preexisting differences.
goal is to establish
Outcomes such as motivation,
causality, not just performance, or satisfaction are
Measurement
correlation. measured after exposure to the
manipulation.
Exercise
• Gender shapes the baseline evaluation of
Factor Levels Purpose
the leader and conditions how motive
information is interpreted. Male
Tests gender
Leader (Daniel) /
• The core logic is that the same stereotypes and
gender Female
transgressive act can be interpreted role congruity.
differently depending on stated motive and (Daniela)
on the actor’s social identity. Pro-self / Tests how
• Give scenario for male vs female leader, Leader Pro- perceived intent
pro self vs pro other unethical behavior motive organizatio shapes moral
nal judgment.
You are an HR business partner in a multinational firm. The annual talent review cycle is
concluding. Leadership has mandated that each unit must categorize 20 percent of
employees as top performers to determine bonuses and promotions. However, your unit’s
results are average, and few employees clearly qualify.
Your manager calls you to discuss the final ratings before submission to corporate HR.
Condition 1 – Male Leader, Pro-Organizational Motive
Your manager, David Sharma, says that several senior leaders have expressed concern
that the unit looks weak compared to others. He explains that if your team submits lower
ratings, the unit’s overall credibility will drop, potentially leading to budget cuts and fewer
development opportunities next year.
David asks you to slightly raise the performance scores of a few borderline employees so
the unit aligns with company benchmarks. He notes that “this is in everyone’s best
interest” and will protect the team’s standing during the upcoming review.
Condition 3 – Male Leader, Pro-Self Motive
Your manager, David Sharma, tells you that his own bonus depends partly on how the
unit’s performance rankings compare with others. He mentions that his boss expects to
see a stronger profile this year.
David asks you to slightly raise the performance scores of a few borderline employees to
make the team appear stronger. He adds that “this is standard practice” and hints that he
would ensure your efforts are noticed when next year’s compensation decisions are made.
Manipulation checks
1 Perceived leader gender (binary check).
2 Perceived motive (two items, one probing perceived benefit to manager, one
probing perceived benefit to team or organization).
3 Perceived severity (single item) to ensure same judged magnitude across
conditions.
References for vignette manipulation checks and best practice (Aguinis and
Bradley 2014).
Dependent variables and recommended items (7 point Likert, strongly
disagree to strongly agree)
Intent to comply (I would follow the manager’s request) (Umphress and
Bingham 2011).
Perceived unethicality (The requested action is ethically problematic) (Lee et al.
2019).
Attribution of motive (The manager is acting to benefit themself) and (The
manager is acting to benefit the team) (adapted from Malle 1999).
Leader evaluation composite (trustworthiness, integrity, competence) (Brown,
Treviño, and Harrison 2005).
Reporting intention (I would report or discuss this with HR or compliance)
(Detert and Treviño 2010).
You are an HR business partner reviewing expense reimbursements
for your department. During a routine audit, you notice several
borderline expense claims that require manager approval before
submission. Your manager calls you into a private meeting to finalize
these reimbursements.
Condition 1 – Female Leader, Pro-Self
Maya Rao adjust her personal expenses in a professional
travel reimbursement.
Condition 2 – Female Leader, Pro-Other
Maya Rao adjust a colleagues personal expenses in a
professional travel reimbursement. She emphasizes that
approving this adjustment will help the employee.
HR Analytics
Session 12
Anshul M
GIM
Cynet Systems
What challenges did Cynet face in the shift to virtual work mode?
1. Difference between covid and non-covid context of work
2. Ensuring that all employees had adequate technology infrastructure in place
for them to be able to work remotely.
3. Orientating employees on working in a remote environment. The average
work experience of Cynet team members as of 2021 was only thirty-five
months, and about 18 per cent of employees (199 of 1,110) were recent
graduates.
4. Giving employees time to adapt to a remote working environment.
5. Team leaders managed employee performance data in a decentralized
manner. Developing a strategic view of how HRA would contribute to
business success. Functioning in a dynamic environment, HR at Cynet was
still trying to determine this.
6. Adjusting to the fact that the overall adoption of HRA was in the early
stages. The key decision-makers at Cynet were unsure how to build a
predictive analytical dashboard, leaving the team leaders to micromanage
Was Cynet ready to adopt people
analytics for managing performance
remotely?
• Technological infrastructure?
No centralized database, limited
existing metrics
• Organizational factors?
The type of work they are into
External factors/covid shock?
US economy, sudden push and pull of
demands-
Individual factors?
Freshers, no existing analytics
• Help Sharma develop a report that can be presented to Cynet
management about the recruiters' performance based on the
data from 2019 to 2021. For example, what was the total
headcount growth of Cynet recruiters over the three years?
Calculate some statistics and plot one or more graphs to
investigate whether performance has improved or declined over
this period and by how much. The report can be divided into the
following parts:
• Summary overview of the data
• Analysis of the data across clients served, and onboarding
numbers
• Recommendations and next steps
-Look at the location data
-Look at total data from client served by location
-performance of recruiter?
How can Cynet leverage people anaiytics with the
existing data to build deeper insights about
the recruiters' performance?
• Who are top performer’s here?
• Are inactive people of our interest?
1. Given the data available, could Cynet
capture all issues that affect recruiter
performance using quantitative data only?
What qualitative dimensions should Cynet
measure in this regard?
2. How could Cynet develop its people
analytics for strategic decision-making? What
do firms
require to achieve digital transformation to
inform strategic choices?