Primary and secondary effects Preliminary results

1
Primary and secondary effectsPreliminary results

• Michelle Jackson

2
Transition to A level workClass, 2001
3
Transition to A-level workRelative importance of
secondary effectsClass, 2001
4
Transition to A level workSex, 2001
5
A note on the method of calculating relative
importance

• The Erikson et al. method of evaluating the
  importance of secondary effects only works when
  the initial odds ratio is reduced in size by
  taking account of secondary effects, and when the
  direction of the odds ratio is the same
• In this case, men have an estimated transition
  rate of 59, women 62. If we take a
  counterfactual case with mens performance but
  womens transition rates, we find a total
  transition rate of 58 Similarly, if we take
  womens performance but mens transition rates,
  the total transition rate would be 63.
• Where a counterfactual odds ratio is larger or
  runs in the opposite direction, it makes no sense
  to calculate relative importance in the way that
  we have done previously.

6
Transition to A-level work Relative importance
of secondary effectsSex, 2001
7
Transition to A level workEthnicity, 2001
8
Transition to A-level work Relative importance
of secondary effectsEthnicity, 2001
9
Transition to A-level workRelative importance of
secondary effectsEthnicity, 2001
10
Transition to A level workClass, 2001
11
Transition to degree (given A-level
performance)Class, 2001
12
Transition to degreeRelative importance of
secondary effectsClass, 2001
13
Transition to A level workSex, 2001
14
Transition to degree (given A-level
performance)Sex, 2001
15
Transition to degreeRelative importance of
secondary effectsSex, 2001
16
Transition to A level workEthnicity, 2001
17
Transition to degree (given A-level
performance)Ethnicity, 2001
18
Transition to degreeRelative importance of
secondary effectsEthnicity, 2001
19
To do

• Develop a way of judging the significance of
  relative importance ratios, so that we can say
  whether there are real changes over time, and
  whether there are real differences between groups
• Break down the ethnic groups (and perhaps
  classes) to smaller groups, which should be
  possible by pooling several surveys
• Complete analyses for entire time series