Measuring Offshorability: A Survey Approach

1
Measuring Offshorability A Survey Approach

• Alan Blinder and Alan Krueger
• Princeton University
• October 4, 2008

2
Background

• Offshorability is potentially important but hard
  to measure, maybe even hard to define.
• Blinder (2007) tried, making judgments by
  occupation.
• This paper seeks to develop individual-based
  data.
• A key question for us Can survey questions be
  designed to track offshorability over time?

3

• Westat External Coding (Case I)
• Based on 2003 National Assessment of Adult
  Literacy
• Stratified Random Sample of 3,000 cases
• Information

Codes
4
(No Transcript)
5
NAAL 2003 Results Externally-Coded Offshorability

Assessment
Percent 1 not
offshorable.... 71.9 2
offshorable with considerable difficulty ...
4.8 3 mixed or neutral..
6.2 4 offshorable with some
difficulty. 5.3 5 offshorable
with minor or no difficulty.. 11.8


N 2,985
6
PDII Survey

• Conducted by Westat English and Spanish
• Start with CPS questions
• Autor Handel, Blinder, Kleiner, Krueger, Blau
  Kahn, Hall, Freeland helped design new questions
  on occupational licensing, experience, etc.
• Focus group tests
• Codebook, questionnaire, and documentation on
  Kruegers web page under PDII. Data will be
  posted there soon.

7
More on PDII Survey

• Universe Individuals age 18 or older in the
  labor force
• 2,513 eligible individuals were interviewed June
  5-July 20, 2008
• Asked about last job for those not currently
  employed
• Random digit dial sampling design
• One respondent randomly selected from eligible
  household to complete the survey
• Up to 15 callbacks made to try to elicit
  responses.
• 28 of sampled eligible households agreed to
  participate in the screening questions, and 64
  of the selected individuals in screened
  households completed the questionnaire.
• Response rate was 17.9, using AAPOR definition 3
• Sample weights primarily match the marginal
  distributions of the Current Population Survey by
  sex, age, education, census region, urbanization,
  race, Hispanic ethnicity, employment status, and
  class of employer

8
PDII Survey Data Main Self-Reported Question
We later combine 2 and 4 to make a 3 point scale,
and re-norm so that 1 is least offshorable and 3
is most.
9
Partial Validation of Self-Reported
Offshorability (Q27)

• (1) Asked 197 subjects the reason why job could
  be done remotely or requires physical presence.
  In free-form responses
• -- 11.7 gave responses suggesting they
  misunderstood
• the question. (Example Someone said their work
  required physical presence because it comes
  through the computer queues that are on a
  mainframe at the job location.)
• -- The error rates were about the same for
  offshorable jobs (10.8) and non-offshorable jobs
  (11.9)
• Self-reported offshorability was higher for jobs
  that did not require face-to-face contact.
• (3) Puzzle Self-reported offshorability is not
  higher for jobs where people make things than for
  jobs where they provide services. (This is in
  contrast to the externally-coded offshorability
  variable.)

10
Job Features Related to Offshorability (1)
11
Job Features Related to Offshorability (2)
Q31. Now think about the work you do
face-to-face with others. To what extent is it
possible for you to do that work without being
physically present? By that I mean doing the work
at a remote location and then delivering it by
mail, by telephone, by sending it over the
Internet, and so on. Would you say all of the
work could be done that way, most of the work, a
little of it, or none at all? ALL OF THE WORK
1 3 MOST OF
THE WORK 2 13 A
LITTLE OF THE WORK 3
28 NONE AT ALL 4
43 Not asked because they were
skipped 12
12
PDII Survey Data Westat Coders Assessments of
Offshorability based on Job Tasks and Business

Assessment
Percent 1 not
offshorable.... 68.3 2
offshorable with considerable difficulty ...
8.3 3 mixed or neutral..
6.3 4 offshorable with some
difficulty. 6.3 5 offshorable
with minor or no difficulty.. 10.8

• We will also combine 1 and 2, and 4 and 5 to
  create a 3 point scale.
• Note Slight drop from 71.3 not offshorable in
  NAAL in 2003
• to 68.3 in PDII in 2008.

13
PDII Sample Comparison of Externally-Coded and
Self- Reported Offshorability, Collapsing Both to
3-Point Scales
WESTAT CODED
1 2 3
Total -------------------------------------------
----------- 1 55.4 2.9
8.9 67.3 -------------------------------
----------------------- 2 10.3
1.1 2.5 13.9
-----------------------------------------------
-------- 3 10.9 2.2
5.7 18.8 -----------------------------
--------------------------
76.6 6.3 17.1 100.0
Q27 SELF-REPORTED
N2,247
Expected Agreement
Agreement Kappa Std. Err. Z
ProbgtZ -------------------------------------------
----------------------------------------- 63.3
56.1 0.1654 0.0158
10.48 0.0000
14
Offshorability by Industry
15
Findings related to other PDII projects

• Licensure Jobs that require an occupational
  license are less likely to be reported as
  offshorable.
• Routinizablity No difference in offshorability
  between jobs in which workers carry out short,
  repetitive tasks more than versus less than ½
  the time for externally coded data, less (?)
  offshorability for Q27.

16
PDII Sample Ordered Probit Model to Predict
Offshorability Dependent Variable (1, 2 or 3)
rising in offshorability
17
Wage regressions with Offshorability on RHS
18
Conclusions

• Self-rated and externally-coded offshorability
  yield similar marginal distributions ? similar
  aggregate measures of offshorability.
• These estimates are strikingly similar to those
  in Blinder (2007).
• But the joint distribution is not diagonal ? the
  two seem to measure different things.
• Some support for validity of each measure
• Offshorable jobs may be concentrated more in the
  higher end of the skill distribution.
• Question Is it worthwhile to code occupation and
  industry information from earlier CPSs and/or
  Censuses to track trends in offshorability?