Concepts to be included (causal) hypothesis Bivariate Explanation/confounding (effect of a ‘third’ variable) Confounding variable Specification/interaction/modification Modifier (interaction variable) Footer text: to modify choose ‘Insert’ (or ‘View’ for office 2003 or earlier) then ‘Header and Footer’ 4/29/2019

The effects of third variables Henk van der Kolk

Aim Understanding the potential effect of ‘third’ variables (spurious relationships) I discuss two effects: Explanation/Confounding Specification/ Interaction/ modification

Two examples Why are more babies born (per capita) in some municipalities than in others? (to illustrate ‘confounding’) Why do some people spend more on holidays than others? (to illustrate ‘interaction’)

Three aspects of causality Cause precedes the consequence in time (correct time order) Consequence occurs less often if the cause is absent (association) No common cause of cause and consequence (more generally: no third variable affecting the relationship) (no spurious relationship)

Spurious relationships Relationships can be ‘spurious’ or seriously biased because of … Explanation / confouding Specification/ interaction/ modification. ‘Third variable bias’. De puntjes vallen nu precies op de volgende regel. De zin net een woordje langer of korter maken

Example I: confounding In municipalities with a large number of storks per capita, the number of children per capita is relatively high, while in municipalities with a small number of storks, the number of babies is relatively low”

The expectation in a model Number of storks Number of babies

The expectation in a graph Babies Storks

Testing the causal relationship Babies causing storks? (correct time order) Correlation? (we found an association) What would be the ‘theoretical argumentation’ about why this is still NOT a causal relationship?

Degree of urbanization Confounding in a model Degree of urbanization Number of storks Number of babies

Confounding in a graph Not urbanized (1) Babies Urbanized (3, low) Urbanized (5, medium) Urbanized (7, high) Urbanized (10, highest) Storks

Example II: interaction/Specification ‘Households with a high level of income, will spend relatively more on holidays, than people with a low level of income.’

Holiday spending Income Units: households Variables: income & holiday spending Sign: positive

Spending for holidays Income

Testing the causal relationship Holidays causing income? (correct time order) Correlation? (we found an association) What would be the ‘theoretical argumentation’ arguing why this is NOT a simple causal relationship?

Willingness to go on holiday Interaction in a model Modifier variable Willingness to go on holiday Income Holiday spending

Interaction in a graph Wants to go Spending for holidays Does not want to go Income

This microlecture Understanding the potential effect of ‘third’ variables Confounding (aka ‘Explanation’) Interaction (aka ‘Modification’ & ‘Specification’)

Footer text: to modify choose ‘Insert’ (or ‘View’ for office 2003 or earlier) then ‘Header and Footer’ 4/29/2019

Images used Slide 7: https://pixabay.com/en/storchennest-stork-couple-storks-362889/ Slide 13: https://pixabay.com/en/clubs-beach-baltic-sea-holiday-518192/