Summary Missing Data and causal effects
Inhoud
Missing Data: HC 1................................................................................................................................... 2
Literature week 1 .................................................................................................................................... 7
Book chapter 1.1 + 1.2......................................................................................................................... 7
Missing data HC 2 .................................................................................................................................... 9
Literature week 2 .................................................................................................................................. 17
Book chapter 1.3 + 1.4....................................................................................................................... 17
Missing data HC 3 .................................................................................................................................. 25
Literature week 3 .................................................................................................................................. 30
Rubin chapter 1 ................................................................................................................................. 30
Missing Data HC 4.................................................................................................................................. 33
Literature week 4 .................................................................................................................................. 38
Rubins: Chapter 5 .............................................................................................................................. 38
Rubins: chapter 6............................................................................................................................... 39
Missing data HC 5 .................................................................................................................................. 41
Literature week 5 .................................................................................................................................. 46
Book chapters: 2.1 + 2.3 + 2.4 + 2.7 + 2.8 + 3.1 + 3.4 ........................................................................ 46
Missing Data HC 6.................................................................................................................................. 57
Literature week 6 .................................................................................................................................. 65
Book chapter 8 .................................................................................................................................. 65
Rubins chapter 6.4 + 6.5 .................................................................................................................... 66
Missing data HC 8 .................................................................................................................................. 68
1
,Missing Data: HC 1
Missing values
Missing values are those values that are not observed → Values do exist in theory, but we are unable
to see them (e.g. non-response → not being home, not answering, or measurement breaks down
etc. ). In this course the main focus is non-response. There are different kind of non-response:
Types of non-response
1. unit non-response: no observed response at all for a case → people are in total not
participating
2. item non-response: some, but not all, responses are missing for a case → some data we have
but some not.
➔ Unit nonresponse refers to the complete absence of an interview from a sampled
household whereas item nonresponse refers to the absence of answers to specific
questions in the interview after the sampled household agrees to participate in the
survey. OR Unit nonresponse in a survey occurs when an eligible sample member fails to
respond at all or does not provide enough information for the response to be deemed
usable (not even as a "partial completion"). Unit nonresponse can be contrasted to item
nonresponse (missing data) wherein the sample member responds but does not provide
a usable response to a particular item or items.
Furthermore You can classify missing values in three groups:
1. Unintentional: Missing values that should have been observed
2. Intentional : Missing values that should not have been observed (e.g. → Q1: do you have
children? ; Q2: what is the age of your children? → if the answer on Q1 is ‘no’ than the
answer on Q2 is supposed to be missing
3. Deductive missings: Missing values whose true value can be deduced from the observed data
(e.g. → when you don’t have information about the BMI, you can calculate it with someone’s
height and weight.)
Sampling: people are not participating and are supposed to do not
2
,Branching / matrix sampling: people are not filling in an answer and are supposed to
Refusal / self-selection: people are not participating that are supposed to do: can lead to a selection
bias
Skip question / coding error: questions are not answered that are supposed to be answered, maybe
because of a typo.
Why missing values?
Death, dropout, refusal,bad luck → USB stick lost etc ..
But more importantly: If not all necessary information is captured, our inference may be wrong. This
can be due to errors with respect to: sampling (does sampling match the research goals?), coverage
(is the target population the same as the targeted population?), non-contact (unable to reach the
respondent), incompetence of the researcher or just refusal.
Bold statement 1
EVERYTHING IS A MISSING DATA PROBLEM → there is always some source of information missing
3
, Experiment example: impossible because you want to compare above vs above
➔ Solution: randomization
Illustration of the problem
Example 1
➔ The same holds for a x2 regression, variance etc ..
➔ Furthermore: Because we have a smaller observed set (when compared to the incomplete
set) there is lower statistical power (founding an effect when there is = linkerbovenhoek).
and a smaller effect size (R2).
4
Inhoud
Missing Data: HC 1................................................................................................................................... 2
Literature week 1 .................................................................................................................................... 7
Book chapter 1.1 + 1.2......................................................................................................................... 7
Missing data HC 2 .................................................................................................................................... 9
Literature week 2 .................................................................................................................................. 17
Book chapter 1.3 + 1.4....................................................................................................................... 17
Missing data HC 3 .................................................................................................................................. 25
Literature week 3 .................................................................................................................................. 30
Rubin chapter 1 ................................................................................................................................. 30
Missing Data HC 4.................................................................................................................................. 33
Literature week 4 .................................................................................................................................. 38
Rubins: Chapter 5 .............................................................................................................................. 38
Rubins: chapter 6............................................................................................................................... 39
Missing data HC 5 .................................................................................................................................. 41
Literature week 5 .................................................................................................................................. 46
Book chapters: 2.1 + 2.3 + 2.4 + 2.7 + 2.8 + 3.1 + 3.4 ........................................................................ 46
Missing Data HC 6.................................................................................................................................. 57
Literature week 6 .................................................................................................................................. 65
Book chapter 8 .................................................................................................................................. 65
Rubins chapter 6.4 + 6.5 .................................................................................................................... 66
Missing data HC 8 .................................................................................................................................. 68
1
,Missing Data: HC 1
Missing values
Missing values are those values that are not observed → Values do exist in theory, but we are unable
to see them (e.g. non-response → not being home, not answering, or measurement breaks down
etc. ). In this course the main focus is non-response. There are different kind of non-response:
Types of non-response
1. unit non-response: no observed response at all for a case → people are in total not
participating
2. item non-response: some, but not all, responses are missing for a case → some data we have
but some not.
➔ Unit nonresponse refers to the complete absence of an interview from a sampled
household whereas item nonresponse refers to the absence of answers to specific
questions in the interview after the sampled household agrees to participate in the
survey. OR Unit nonresponse in a survey occurs when an eligible sample member fails to
respond at all or does not provide enough information for the response to be deemed
usable (not even as a "partial completion"). Unit nonresponse can be contrasted to item
nonresponse (missing data) wherein the sample member responds but does not provide
a usable response to a particular item or items.
Furthermore You can classify missing values in three groups:
1. Unintentional: Missing values that should have been observed
2. Intentional : Missing values that should not have been observed (e.g. → Q1: do you have
children? ; Q2: what is the age of your children? → if the answer on Q1 is ‘no’ than the
answer on Q2 is supposed to be missing
3. Deductive missings: Missing values whose true value can be deduced from the observed data
(e.g. → when you don’t have information about the BMI, you can calculate it with someone’s
height and weight.)
Sampling: people are not participating and are supposed to do not
2
,Branching / matrix sampling: people are not filling in an answer and are supposed to
Refusal / self-selection: people are not participating that are supposed to do: can lead to a selection
bias
Skip question / coding error: questions are not answered that are supposed to be answered, maybe
because of a typo.
Why missing values?
Death, dropout, refusal,bad luck → USB stick lost etc ..
But more importantly: If not all necessary information is captured, our inference may be wrong. This
can be due to errors with respect to: sampling (does sampling match the research goals?), coverage
(is the target population the same as the targeted population?), non-contact (unable to reach the
respondent), incompetence of the researcher or just refusal.
Bold statement 1
EVERYTHING IS A MISSING DATA PROBLEM → there is always some source of information missing
3
, Experiment example: impossible because you want to compare above vs above
➔ Solution: randomization
Illustration of the problem
Example 1
➔ The same holds for a x2 regression, variance etc ..
➔ Furthermore: Because we have a smaller observed set (when compared to the incomplete
set) there is lower statistical power (founding an effect when there is = linkerbovenhoek).
and a smaller effect size (R2).
4
