Statistics: An Introduction

1
Statistics An Introduction

• LIR 832
• Class 1
• January 8, 2007

2
Topics of the Day

• 1. Some really nice pictures (graphical display
  of quantitative data and more)
• 2.  Why do we teach statistics other than to
  make your life miserable for a semester?
• 3. What is in it for me as a future
  HR/LR practitioner?
• 4.  Fundamental issues in statistics (what
  really matters)
• 5. The structure of the course

3
The Use of StatisticsClassic Examples

• Tufte The Visual Display of Quantitative
  Information
• For a successful technology, reality must take
  precedence over public relations, for Nature
  cannot be fooled (Richard Feynmans conclusion
  on the explosion of the space shuttle).

4
The Use of StatisticsClassic Examples

• Data map John Snow (1854) deaths from cholera
  in Central London
• Before knowledge of bacterial sources of illness
  (Holmes father in 1847 Louis Pasteur later on).
• Deaths are dots xs are water pumps
• Deaths are clustered around the Broad St. Pump
• Removal of handle ended an epidemic which killed
  more than 500

5
The Use of Statistics Classic Examples

• Data map National cancer rates by county
• From darkest (in highest decile of cancer rates)
  to lightest (lower than US as a whole).
• High death rates from cancer in the northeast
  part of the country and around the Great Lakes
  (High levels of air pollution and dense
  concentration of industry).
• Low rates in an east-west band across the middle
  of the country.
• Higher rates for men than for women in the south,
  particularly Louisiana (cancers likely caused by
  occupational exposure from working with asbestos
  in shipyards).
• Can you find the counties which are downwind from
  the Nevada test range?
• Can you find the central locations for the
  chemical industry in the US?

6
The Use of Statistics Classic Examples

• Data Map Space and Time - Charles Minard (1861)
  - Napoleons March
• Width of line varies continuously with size of
  the army.
• The line establishes the longitude and latitude
  of the army.
• The lines show the direction of movement of the
  army.
• The location of the army with respect to certain
  dates is marked.
• The temperature along the path of march is
  marked.

7
The Use of StatisticsClassic Examples

• Computer Graphics Space and Time
• Concentration of Pollutants over L.A. July 22,
  1979
• Two dimensional surface 6 south California
  counties
• Nitrous oxides power plants, refineries
  vehicles
• Refineries and Kaiser Steel produce post midnight
  peaks.
• Traffic and power plants produce daytime peaks
• Carbon monoxide
• Reactive hydrocarbons

8
The Use of StatisticsClassic Examples

• Election Maps Difficulty in Portraying
  Information Accurately and making your point.
  (http//www-personal.umich.edu/mejn/election/)

9
The Use of StatisticsClassic Examples

• Tuftes Principles of Graphic Excellence
• The efficient communication of complex
  quantitative ideas
• Show the data
• Avoid distorting what the data have to say
• Encourage the eye to compare different pieces of
  data
• Make large data sets coherent
• Induce the viewer to think about the substance
  rather than about methodology, graphic design,
  the technology of graphic production, or
  something else

10
The Use of StatisticsTruck Driver Retention

• Factors Affecting Over-the-Road Truck Driver
  Retention A More Traditional Application of
  Statistics to a Complex Relationship

11
The Use of StatisticsTruck Driver Retention

• Background
• Ongoing shortage of truck drivers makes trucking
  firms very concerned, at least rhetorically,
  about driver retention
• Have excellent data on drivers from a survey of
  truck drivers, would like to sort out factors
  affecting driver retention so firm policy can
  focus on those factors
• Problem, retention is multi-causal, many factors
  are likely to affect the retention of truck
  drivers and we need an approach that allows for
  all of these affects.

12
The Use of StatisticsTruck Driver Retention

• It is always good to start an inquiry with a
  little theory. This sets a question or questions
  that we structure our inquiry around.
• The following from Freeman and Medoff Two Faces
  of Unionism

13
The Use of StatisticsTruck Driver Retention

• Monopoly face unions raise wages and improve
  benefits
• Exit-voice face.
• Typical means of employees registering
  dissatisfaction with a job is to quit and find a
  new job.
• Unions provide employees with an alternative
  route voice
• Improve communications because employees are
  protected against bad consequences of
  communicating their views to management
• Allow employees a means to communicate and decide
  on issues among themselves rather than being
  mediated by management. Employees rather than
  management decide on hard issues such as the
  allocation of benefits
• Solves public goods problem at work

14
The Use of StatisticsTruck Driver Retention

• Lower quit rates and longer tenure are a
  potential source of advantage to organized firms
  as they
• Save hiring and training costs
• Have greater depth of human capital
• Research on quits and employee tenure shows a
  strong positive association between tenure (years
  of service with employer) and unionism and a
  strong negative association between unionism and
  quits.

15
The Use of Statistics Truck Driver Retention

• This might be explained by the union voice
  effect but it might also be an example of the
  monopoly face (syllogism)
• Unions raise wages and increase benefits
• All else constant, employees tend to stay with
  firms which provide better wages and benefits
• To be better assured of union voice effects we
  need to distinguish the monopoly face of unions
  on compensation from that of voice
• Consistent with the monopoly argument, Delery
  finds only a compensation effect, no distinct
  union effect

16
The Use of StatisticsTruck Driver Retention

• Current Research draws on a UMTIP survey of truck
  drivers.
• Interview 1,000 drivers in truck stops between
  1997 and 1999
• Includes data on tenure and quits along with
  union membership and compensation.
• Consider the descriptive statistics (abbreviated)

17
(No Transcript)
18
The Use of StatisticsTruck Driver Retention

• Build a series of models that look at the months
  spent with their current employer (tenure)
• Dont have pre-quit information on those who quit
  their jobs in the last year).
• Models (working from simplest to most complete)
• Model 1 Tenure models with extensive controls
• Controls serve to eliminate the effects of
  factors which would otherwise confuse our
  estimates, such as personal characteristics
  which might affect tenure (age, race, gender
  ...), segment of the industry, size of the firm,
  characteristics of the work.
• Model 2 Add Union Measure
• Coefficient on union membership is 38.78
  interpreted as indicating union members stay with
  firm an additional 39 months.

19
(No Transcript)
20
The Use of StatisticsTruck Driver Retention

• Model 3
• Add a measure of weekly pay. The coefficient on
  union membership is not the earnings (monopoly)
  effect as we have removed effects related to
  weekly earnings.
• Members remain an additional 36 months, not much
  effect
• Model 4
• Allow for the effect of benefits including paid
  days off, employer provided health insurance,
  pensions and deferred compensation
• Union effect declines to an additional 22 months

21
The Use of Statistics Truck Driver Retention

• Conclusions
• Union membership has a strong effect on employee
  retention. While part of this effect is due to
  unions improving wages and benefits, even with
  controlling for such effects, unions continue to
  be associated with longer employee retention.
• Time off work is also very important to driver
  retention as is earnings.

22
Why Quantitative Methods?

• What will we learn?
• Master fundamental knowledge about construction
  and application of statistical models.
• Develop, operationalize and interpret models of
  social interaction using modern statistical
  software.
• Learn to evaluate and critique others research.
• In essence, become knowledgeable users of
  statistical analyses.

23
What Issues Does Statistics Address?

• Human beings are very good storytellers.
• Human beings have always been very good at
  developing stories which explain the world out
  of small amounts of information. This behavior
  may have been necessary for survival when early
  man competed for food with large predators, but
  it often leads us to misunderstandings about
  causal relations.

24
What Issues Does Statistics Address?

• Tom Peters In Search of Excellence Lessons
  from America's Best-Run Companies
• How 10 firms became top performers. Exciting
  reading with many important insights into
  successful management. ATT, IBM, Digital
  Equipment, 3M, Allen-Bradley, Delta Airlines
• Fortune magazine returns three years later, half
  of the firms are no longer top performers

25
What Issues Does Statistics Address?

• Beardstown Ladies
• Successful investment club in Ohio. Produces of
  book of investment tips with recipes. Some
  problems later on with how they figured their
  profits but lets put that aside. Their claim to
  fame was that they out guessed the stock market
  ten years in a row. Did this reflect brilliant
  thinking on their part, or might it simply be
  luck (change or random event)?
• Supppose in 1980 there were 1000 womens
  investment clubs in Ohio. Each year we would
  expect ½ of those clubs would do better than the
  stock market and one half would do worse. How
  many clubs would have a record of straight wins
  in the 1980s?

26
What Issues DoesStatistics Address?
27
What Issues Does Statistics Address?

• Stories are essentially anecdotes, interesting
  and potentially insightful but its difficult to
  separate what is useful from what is bullshit.
  Much of what we consider theory in social and
  behavior science, whether it be economic theory,
  psychological theory, sociological theory,
  management theory, physics, etc.

28
What Issues Does Statistics Address?

• Statistics provide a method of examining these
  stories to determine if they are consistent with
  the facts (data) generated from a large number of
  cases.
• For example, the questions we might be interested
  in might be
• Does a particular absence policy actually reduce
  absences?
• What is the response to a pay for performance
  system?
• How does greater workforce diversity influence
  plant level performance?

29
What Can We Do With Statistics?

• Compactly summarize large bodies of data
• Using measures of central tendency, dispersion
  and probability distributions, we can compactly
  describe and understand these large bodies of
  data.
• CPS file with 150,000 observations on earners
• Compustat file with annual data on 1,000s of
  firms at the divisional level
• Personnel files from medium or large size
  corporations.

30
What Can We Do With Statistics?

• Determine if there are meaningful relationships
  in the data
• Test theories or ideas about social or other
  inter-relationships
• An attendance award program will reduce
  absenteeism
• Piece rate systems will increase output but
  quality will suffer (Dodge Brothers machining
  plant in Detroit in 1904 for example)
• Increases in the minimum wage reduce employment
• Training programs improve output
• What is a theory? (evolution v. intelligent
  design)

31
What Can We Do With Statistics?

• Determine the magnitude of the relationship
  Answer the essential question How Big
• If you are going to calculate the ROI on a
  training program you need to know the magnitude
  of the effect of that program. So you will want
  to be able to answer questions such as
• Following training program X, productivity rose
  by Y
• If a firm is going to invest in a program, it
  need to know the rate of return and this will, in
  turn, be determined by the improvement in
  productivity.
• An A increase in the minimum wage is associated
  with a B decline in teenage employment.
• Piece rate workers produce H more output than
  hourly workers.

32
What Can We DoWith Statistics?

• This is all very nice, what is in it for you as
  an IR/HR professional?
• IR/HR students do not, typically, believe that
  numbers are their friends.
• Alas, HR Managers are expected to use numeric
  and statistical information to understand and
  guide their decisions.
• As HR moves from a transactional to strategic
  position within the firm, HR managers are more
  and more expected to use numeric and statistical
  methods to evaluate operations and guide their
  decisions.
• Organizational HR performance is monitored using
  HR metrics. If correctly chosen these metrics
  can provide a compact summary of units
  performance.

33
What Can We Do With Statistics?

• A Few HR Metrics
• Number of interviews to hires
• Total recruiting cost per hire
• Hiring manager satisfaction
• Turnover Rate
• Turnover Cost
• Absence Rate
• Health Care Cost per Employee
• HR expense factor
• Human Capital Value Added
• Workers Compensation Cost per Employee
• HR ROI

34
What Can We Do With Statistics?

• Some issues in using these metrics
• What do these measure?
• What is a good performance?
• Are deviations from good performance due to
  problems or chance?
• What are the sources of good or bad
  performance?

35
What Can We DoWith Statistics?

• With the availability of HR metrics, it become
  possible to use descriptive and analytic
  statistics to evaluate programs.
• Consider a program to control health care costs.
  You are going to be interested in some relatively
  simple measures such as whether there was a
  reduction in direct health care costs. You will
  also be interested in determining whether there
  are indirect costs such as increased absenteeism,
  lower employee satisfaction, increased turnover
  and whether there is a change in employee
  behavior or simply cost shifting.

36
What Can We DoWith Statistics?

• You will regularly be presented with reports and
  memos incorporating numeric and statistical
  materials. You needed to understand and evaluate
  the work of others
• You hire a consultant to suggest or evaluate a
  program. You need to be able to understand and
  interpret what they have done both to determine
  the quality of the work, to be able to ask good
  questions, and to reach your own conclusions
  about the report.
• Example EEOC and OFCCPs standards for
  establishing a pattern and practice of
  discrimination

37
What Can We DoWith Statistics?

• You should be facile with statistical measures
  and data to be able to play with professions for
  which is required knowledge. You can also shine
  relative to your peers if you are the one who
  does the statistical work and drafts those
  reports.

38
Fundamental Issues in Statistics

• The world is multi-causal, meaningful models need
  to reflect multiple sources of causation.
• There are many random elements in the outcomes we
  are concerned with. Simple observation is not
  enough to reveal underlying relationships. We
  need multiple observations to be able to
  establish the presence of a relationship.
• Why anecdotes are suspect.

39
Fundamental Issues in Statistics

• We use samples to learn about populations
• We seldom observe the populations we want to know
  about. Because we have to use samples, we engage
  in inference from samples to populations.
  However, because of sampling variability, samples
  are not little mirror images of the population of
  interest. Given that samples are imperfect
  replications of populations, we have to use
  techniques such as hypothesis testing to
  determine if statements about populations are
  reasonable given our observed population.

40
Fundamental Issues in Statistics

• Few events have only one or two causes. As we
  want to avoid reductionist approaches, our
  methods must allow for with multiple causation.
• The foundation of model building is not statical
  but theoretic and practical knowledge of an
  issue.
• Evaluation of the usefulness of models then rests
  on both statistical knowledge and broader
  understanding of an issue. Good statistical
  technique is a necessary but not sufficient
  condition for building a useful model.
• Toward the end of the course we will evaluate
  literature using statistics so that we can bring
  all of the diverse elements together.
• Successfully modeling this multi-causal world
  requires careful application of statistical
  technique.
• Traditional course ends with a brief smattering
  of multi-variate statistics, but we need more.

41
The Structure of the Course