Popular in Professional

Medicine

Induction of Social Contagion

Yale University Nicholas A. Christakis

Business

The Allegory of the Favela: The Multifaceted Effects of Socioeconomic Mobility

University of Pennsylvania Leandro S. Pongeluppe

Architecture

Connecting Cities and Nature

University of Virginia Timothy Beatley

Human Rights Law

Human Rights and Corruption: Problems and Potential of Individualising a Systemic Problem

Max Planck Institute Anne Peters

Finance

Chinese Exchange Rate Policy: Lessons for Global Investors

University of California San Diego Michael Melvin

Law

A statistical approach to law school citation rankings

University of Virginia Joshua Fischman

Education

Towards Stronger Causal Claims in Management Research

View more insights in Professional - Business - Management

View more figures from Erasmus University Rotterdam

View the page of Joep Cornelissen

Read the Study

Next in Business

The Making of the “Good Bad” Job: How Algorithmic Management Manufactures Consent Through Constant and Confined Choices

Lindsay Cameron | University of Pennsylvania

Achieving Mutual Understanding Without Saying a Word

Xiao-Ping Chen | University of Washington

Joep Cornelissen addresses widespread concerns about the reliability and strength of many causal claims made in management research.

Image courtesy of interviewee. May 7, 2024

Log-in or Sign-up to Faculti
Currently viewing this subject insight as a guest. You have insight(s) remaining for this month. Login to view 8000+ figures on the platform.

Generating Workspace...

Export as PDF

This video video discusses the significance of understanding causal relationships in management research and social sciences. It emphasizes that establishing clear causal links between events (where condition A reliably leads to effect B) is crucial for informing policy and practice. The speaker identifies three main approaches to studying causation:

1. **Propositional Approach**: This video video method focuses on direct cause-effect scenarios, where researchers hypothesize and test whether a specific condition leads to an effect, generally using controlled interventions.

2. **Configurational Approach**: This video video method counters the propositional perspective by examining the various conditions and processes that might lead to a specific outcome. It acknowledges the complexity of phenomena and looks at multiple potential causes rather than assuming a root cause.

3. **Process Approach**: This video video approach emphasizes the importance of understanding how events unfold over time, focusing on the sequence of events that lead to particular outcomes, thus highlighting the temporality of causation.

The speaker critiques that while each approach has its strengths—such as the propositional method's ability to show cause and effect clearly or the configurational method's comprehensive view of potential causes—they also have limitations, particularly in dealing with the complexity and interdependence of causal relationships.

This video video suggests a "causal triangulation" method, advocating for the use of all three approaches in conjunction, which would compensate for the limitations of each and lead to more robust causal inferences. It argues that leveraging diverse methods will enhance theoretical explanations and improve practical applications in management and social sciences.

Language:

Export as PDF

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.045 --> 00:01.445
Causal relationships,

00:01.555 --> 00:03.805
have been a concern within management research

00:03.805 --> 00:06.365
and across the social sciences for, for a long time.

00:06.985 --> 00:10.445
Uh, and of causal relationship is really the idea that for

00:11.145 --> 00:13.605
the events are happening in the world that we study,

00:13.705 --> 00:16.765
we can really say that one, condition A is seen

00:16.765 --> 00:19.485
to produce a particular effect, b

00:20.145 --> 00:22.245
and that seemed to be important to be able

00:22.245 --> 00:24.405
to establish those causal relationships so

00:24.405 --> 00:25.925
that in no uncertain terms

00:26.025 --> 00:30.365
and beyond basic correlations, we can say a will on

00:31.065 --> 00:32.725
in general lead, lead to B,

00:33.065 --> 00:36.765
either in a very direct deterministic sense that A, always

00:37.405 --> 00:41.215
B always follows from a, the one condition,

00:41.235 --> 00:43.855
or the one effect from, the base condition,

00:44.355 --> 00:46.455
or maybe in a more probabilistic sense

00:46.605 --> 00:50.095
that a on average is more likely to, to lead to B.

00:50.775 --> 00:54.115
So this has been a ma a main preoccupation of,

00:54.225 --> 00:56.715
many social scientists, including management research

00:56.815 --> 00:57.875
for a very long time.

00:58.875 --> 01:01.885
Also, with the perspective that if we, if we're able

01:01.885 --> 01:04.325
to properly identify these causal relationships,

01:04.425 --> 01:07.085
we can also more effectively inform policy

01:07.145 --> 01:10.245
and practice on the things that, managers

01:10.305 --> 01:12.005
or policy makers might, might do.

01:13.305 --> 01:15.915
Yeah. What is interesting is that the topic of causation,

01:16.415 --> 01:19.515
and also questions around causal relationships,

01:19.855 --> 01:24.635
and also the idea that we need to be really certain, be able

01:24.635 --> 01:26.235
to make a robust inference as

01:26.235 --> 01:28.195
to whether a causal relationship is there.

01:28.605 --> 01:30.675
These types of questions haven't really been

01:31.455 --> 01:34.675
su substantively discussed, to any great extent.

01:34.935 --> 01:36.235
So most of the discussion

01:36.345 --> 01:39.155
that has taken place is really at the level of methods

01:39.245 --> 01:43.155
where, typically by importing, techniques

01:43.175 --> 01:45.875
and methods and research designs from econometrics

01:45.935 --> 01:49.835
or from sociology, people have made assumptions about

01:49.835 --> 01:51.195
what causal relationships are

01:51.695 --> 01:54.075
and how they can following these methods,

01:54.255 --> 01:57.035
be properly identified in each of these different contexts.

01:57.545 --> 02:00.165
But what is important to remember here is that,

02:00.345 --> 02:04.485
if you compare, let's say, a experimental design

02:04.515 --> 02:07.205
with a cross case comparative analysis method,

02:07.955 --> 02:11.055
the assumptions that are being made about what causation is,

02:11.125 --> 02:12.695
what a causal relationship is,

02:12.715 --> 02:16.335
and the criteria that are also coming with these methods as

02:16.335 --> 02:20.435
to how to make a causal inference, how to make an inference

02:20.435 --> 02:22.035
that a causal relationship is there,

02:22.415 --> 02:23.555
are actually very different.

02:23.775 --> 02:26.475
Uh, and that discussion hasn't really taken place yet.

02:26.475 --> 02:29.945
And I think it's important to first figure that out

02:30.205 --> 02:33.105
how we generally think about causal relationships,

02:33.415 --> 02:36.865
alongside applying some of these methodological techniques.

02:37.675 --> 02:40.015
Um, yeah, that's, it is a very good question.

02:40.195 --> 02:43.775
So in the main, I've identified

02:43.775 --> 02:46.615
that there are three ways in which we tend to approach

02:47.255 --> 02:48.255
questions of causation

02:48.315 --> 02:51.455
and tend to approach the, the issue of how

02:51.455 --> 02:53.015
to identify cause relationships.

02:53.075 --> 02:56.335
So there's a propositional approach, which is basically

02:56.335 --> 02:58.715
what it says where researchers

02:59.055 --> 03:01.535
Hypothesized that, two events

03:01.835 --> 03:04.215
or, happenings are related

03:04.515 --> 03:06.215
as a simple cause effect scenario.

03:06.395 --> 03:09.645
So one condition, A, directly effects B,

03:09.645 --> 03:12.645
either in a direct way or through some mediating path.

03:12.825 --> 03:16.205
So through c having that ultimate effect, on B.

03:16.905 --> 03:20.445
So you have these very simple if done scenarios where

03:20.965 --> 03:22.805
researchers subsequently do their best

03:22.805 --> 03:24.365
through their research designs

03:25.025 --> 03:28.445
and through their, intervention techniques to make sure

03:28.445 --> 03:29.885
that in a very controlled manner,

03:29.885 --> 03:31.045
they can actually establish

03:31.075 --> 03:35.955
that b in no uncertain terms follows from, from a,

03:35.975 --> 03:37.715
so that's the propositional approach.

03:37.785 --> 03:40.275
Then we have the configurational approach,

03:40.405 --> 03:43.715
which actually emerged as, an antidote

03:43.715 --> 03:45.555
or a challenge to the propositional approach,

03:45.685 --> 03:47.115
where the propositional one

03:47.735 --> 03:51.075
is really looking at a root cause or root condition,

03:51.075 --> 03:54.075
or presupposes a root cause or root condition,

03:54.095 --> 03:57.155
and then looks at the subsequent effect In a very simple

03:57.585 --> 04:01.195
sort of causal scenario, the configurational approach says,

04:01.225 --> 04:03.715
well, actually, what might be better given

04:03.715 --> 04:06.675
that many phenomena are very complex, is to look at

04:07.235 --> 04:09.915
a particular outcome or effect that has materialized

04:10.295 --> 04:13.715
and then look backwards at all the sets of conditions

04:14.375 --> 04:17.395
and processes along, across those conditions

04:17.395 --> 04:19.875
that might have brought that particular effect about.

04:20.605 --> 04:23.305
So it's a very different approach where instead

04:23.305 --> 04:26.385
of looking at the effects of presupposed causes, you look at

04:26.495 --> 04:28.945
what might be the potential causes

04:29.045 --> 04:31.185
or the potential conditions that might have caused,

04:31.885 --> 04:33.305
a particular effect.

04:33.885 --> 04:37.305
And it's as an approach also, compared

04:37.305 --> 04:40.025
to the propositional one, much more alive to

04:40.545 --> 04:44.615
a complex causal field, a whole range of conditions

04:44.795 --> 04:45.935
as potential causes

04:45.935 --> 04:47.415
that might have brought the effect about.

04:47.435 --> 04:51.735
So your frame of reference, is enlarged,

04:51.835 --> 04:53.855
but you also with that look at

04:54.855 --> 04:57.615
multiple conditions alongside one another, and you try

04:57.615 --> 04:59.855
and figure out through your reasoning

04:59.955 --> 05:01.935
and through your probing of what you found.

05:02.805 --> 05:06.145
Uh, so typically these, this is post case comparisons

05:06.145 --> 05:07.145
that you, or,

05:07.145 --> 05:11.145
or comparative research design that you use across cases

05:11.285 --> 05:13.465
or across data sets where you try

05:13.465 --> 05:16.625
and not just identify these potential conditions,

05:16.725 --> 05:20.025
but also counterfactually probe, which of the ones

05:20.025 --> 05:22.145
or which of the sets of conditions alone

05:22.145 --> 05:24.505
or together might have brought this effect about.

05:25.325 --> 05:28.625
So it's a much more systemic form of causation compared

05:28.625 --> 05:30.505
to the, the propositional approach.

05:31.265 --> 05:35.205
And then finally, the process approach, is,

05:35.265 --> 05:36.565
is again, another approach.

05:36.585 --> 05:40.125
So it says, instead of listing sets

05:40.125 --> 05:41.565
of conditions on the one hand

05:41.785 --> 05:44.765
and effects on the other, either in a very simple way

05:44.795 --> 05:46.445
with the propositional approach

05:46.625 --> 05:50.445
or in a more complex way with a configurational approach,

05:51.035 --> 05:52.965
what we should be doing instead is

05:52.965 --> 05:56.365
to follow in a more longitudinal sense how hap

05:56.505 --> 06:00.365
how events play out over time so that we can really see

06:00.515 --> 06:02.205
what is the sequencing of events

06:02.315 --> 06:05.885
that might have materialized in a particular case

06:06.025 --> 06:10.765
or cross cases, and maybe also across larger data sets,

06:10.865 --> 06:13.605
if we have that longitudinal, frame of reference

06:13.605 --> 06:15.445
for those data to see

06:15.995 --> 06:18.605
what effect materialized in that particular case.

06:19.225 --> 06:22.285
So it's a, an approach that takes temporality

06:22.585 --> 06:25.325
and takes the longitudinal perspective very seriously,

06:25.425 --> 06:29.815
and is also alive to the very uncertain probabilistic nature

06:29.835 --> 06:33.015
of the world where sometimes effects get brought about

06:33.325 --> 06:37.615
through certain sequences of events that, that happen.

06:38.435 --> 06:40.655
So they have their own, that approach comes

06:40.655 --> 06:43.575
with its own methodology, it's longitudinal,

06:44.075 --> 06:46.735
but there's also various techniques that you can use to,

06:46.795 --> 06:50.255
to bracket sequences of events to see the transition points

06:50.285 --> 06:53.375
between these, these sequences and to try

06:53.375 --> 06:56.695
and decipher a particular causal process running,

06:56.845 --> 06:58.135
alongside that trajectory.

06:59.225 --> 07:02.195
Yeah, I think, I think the most interesting thing is that,

07:02.575 --> 07:05.835
so each of these approach approaches obviously comes

07:05.835 --> 07:08.675
with its own, prepositions of how to

07:09.515 --> 07:10.795
identify causal relationships.

07:11.375 --> 07:13.995
And it also makes very different assumptions about

07:14.165 --> 07:17.115
where we find these causal relationships and where

07:17.415 --> 07:18.835
and how causation happens.

07:19.695 --> 07:22.635
And one, obvious observation is,

07:22.815 --> 07:25.995
and there's a philosopher called Nancy Cartwright, who

07:25.995 --> 07:28.955
among many others has, has spotted this observation, is

07:28.955 --> 07:32.835
that some approaches are really looking at, so she calls

07:32.835 --> 07:35.595
that the really, really, using causes

07:35.695 --> 07:37.115
or manipulating causes.

07:37.615 --> 07:40.315
So think of the propositional approach where you say, okay,

07:40.375 --> 07:43.185
I'm assuming this is a, a root cause

07:43.285 --> 07:46.985
or a true cause that will bring a certain effect about

07:47.405 --> 07:50.745
either in a deterministic or in a more probabilistic sense.

07:51.255 --> 07:55.025
Then I'm trying to find research designs

07:55.025 --> 07:58.025
and data sets that allow me to manipulate

07:58.025 --> 08:00.665
or in a very controlled manner, see whether that's the case.

08:01.485 --> 08:04.025
And then, so it's really the look looking at the effects

08:04.045 --> 08:08.565
of presumed cause then the other approaches, so the process

08:08.705 --> 08:11.545
and configurational are actually the opposite,

08:11.605 --> 08:13.785
so it's more hunting for causes.

08:13.925 --> 08:17.105
So I know that these types of effects

08:17.165 --> 08:19.625
as phenomena generally occur in the world.

08:20.075 --> 08:24.175
Let's look back and see what might have been,

08:25.275 --> 08:26.775
all the sets of conditions and,

08:26.835 --> 08:29.335
and potential causes, in those terms

08:29.335 --> 08:30.375
that I brought it about.

08:30.435 --> 08:35.255
So it's more hunting for causes, trying to identify and,

08:35.255 --> 08:39.615
and weed out the sets of conditions, that may have,

08:39.715 --> 08:41.575
may have not mattered too much.

08:41.575 --> 08:44.295
Then the sets of conditions that we now can say

08:45.145 --> 08:48.205
are actual causes that produce this, this outcome.

08:48.825 --> 08:52.325
So it's, it's as an approach compared to the using

08:52.345 --> 08:53.725
or manipulating causes

08:53.865 --> 08:56.605
and looking at these very controlled intervention,

08:56.645 --> 08:59.765
experimental based designs that match that.

09:00.075 --> 09:03.405
It's as an approach taking a very different, stance

09:04.105 --> 09:08.125
and looking more at the world out there, so phenomena

09:08.125 --> 09:09.805
that can be observed effects

09:09.825 --> 09:11.285
or outcomes that can be observed

09:11.305 --> 09:12.645
and then be reasoning backwards.

09:13.225 --> 09:15.565
So I think that's a very fundamental distinction,

09:15.585 --> 09:17.645
and it's actually very helpful to think about it

09:17.645 --> 09:19.405
because it also allows you to say

09:20.015 --> 09:22.205
where the one comes with a blind spot.

09:22.385 --> 09:24.045
You know, if I only reason forward

09:24.425 --> 09:28.725
and presume that condition A will BA root cause

09:28.865 --> 09:32.005
and will in most instances lead to B, well,

09:32.125 --> 09:34.205
maybe the reality is a bit more complex

09:34.425 --> 09:37.125
and we have something to learn from the other approach.

09:37.785 --> 09:40.085
Uh, not to say that, you know, I'm not trying

09:40.085 --> 09:41.525
to favor the one over the other,

09:41.665 --> 09:45.285
but, the point with the triangulation idea is to say

09:45.315 --> 09:47.885
that they all come with different vantage points.

09:47.915 --> 09:51.045
They all come with different strengths and limitations

09:51.145 --> 09:54.325
and also different criteria for making, an influence

09:54.325 --> 09:55.685
around the causal relationship.

09:55.785 --> 09:57.685
So let's join them up, let's see

09:57.715 --> 10:00.125
what they collectively offer so

10:00.125 --> 10:02.605
that we get a more robust influence as

10:02.605 --> 10:05.845
to whether causal relationship is there, and can

10:05.845 --> 10:08.525
accordingly theorize better come up

10:08.525 --> 10:11.125
with stronger explanations, but also inform policy

10:11.225 --> 10:13.645
and practice in a, in a much more effective way.

10:14.395 --> 10:18.375
Uh, and to back that up, in the paper, I show how some

10:18.375 --> 10:23.235
of the classic criteria for how we think about, for

10:23.235 --> 10:25.675
how we think about a cause of relationship, so what a cause

10:25.675 --> 10:28.395
of relationship needs to have in order for us

10:28.415 --> 10:31.155
to talk about not just the correlation,

10:31.215 --> 10:35.275
but a course of relationship, how these criteria are not

10:35.995 --> 10:37.075
satisfied by one approach,

10:37.135 --> 10:38.875
but they're more likely to be satisfied

10:38.935 --> 10:40.075
by multiple approaches.

10:40.795 --> 10:43.465
So the propositional one is very good

10:43.535 --> 10:48.265
with its controlled intervention methodology to try

10:48.265 --> 10:51.585
and identify whether a course temporally precedes an effect.

10:51.605 --> 10:55.425
So the course or precedence that is a classic criteria of,

10:55.485 --> 10:58.185
of acausal relationship is, is something that

10:58.375 --> 11:00.105
that approach can do really well,

11:00.745 --> 11:02.405
but it also has a lot of limitations.

11:02.645 --> 11:06.565
I already noted the lack of its ability to deal

11:06.565 --> 11:10.685
with complexity, the lack of its ability to deal

11:10.685 --> 11:12.045
with spurious causation.

11:12.185 --> 11:15.905
The lack of its ability to really beyond doubt show

11:16.015 --> 11:19.505
that there is a dependent relationship between A and B.

11:19.565 --> 11:21.825
So B will necessary

11:21.845 --> 11:24.865
and sufficiently follow from, from a, those are all things

11:24.865 --> 11:26.385
that it struggles with.

11:26.955 --> 11:29.325
Whereas the other approaches then have,

11:29.435 --> 11:31.325
have equity strengths and limitations.

11:31.425 --> 11:34.925
So the process one is very good at, as I described it,

11:34.925 --> 11:39.405
to trace the, the real transitive causal process that goes

11:39.405 --> 11:42.685
through the sets of conditions,

11:42.905 --> 11:46.165
or the events that you've, you've categorized as conditions

11:46.185 --> 11:48.605
or potential causes over time.

11:49.405 --> 11:50.505
So tracing that through.

11:50.605 --> 11:54.605
So the classic notion that we can speak of causation when A,

11:54.675 --> 11:55.805
produce an, an effect,

11:55.805 --> 11:59.205
and B, b produce an effect C, you can trace that

11:59.465 --> 12:03.125
or be much closer to that with the process approach than,

12:03.345 --> 12:04.925
with the other two approaches.

12:05.305 --> 12:08.085
But equally, the process approach also has its limitations,

12:08.145 --> 12:13.085
and again, causal dependence to really showing that b

12:13.345 --> 12:16.725
depends on a, is very difficult to establish in

12:16.725 --> 12:18.085
and of itself by that approach.

12:18.505 --> 12:23.325
And then the configurational approach is very good in trying

12:23.345 --> 12:25.965
to force yourself to look at the world out there,

12:26.265 --> 12:29.245
not limit yourself to one presupposed condition

12:29.265 --> 12:32.205
or limited set of conditions as potential causes,

12:32.225 --> 12:34.825
but really look at the world out there.

12:34.975 --> 12:38.185
What are the regularly occurring connections

12:38.185 --> 12:41.785
between potential causes, conditions as causes,

12:41.885 --> 12:43.745
and the effects that that materialize.

12:44.285 --> 12:49.265
So it has this classic criteria of regularity, in,

12:49.565 --> 12:50.945
in the events that we observe

12:50.965 --> 12:53.825
and the connections between them, which is, which has

12:53.925 --> 12:57.265
for a long time been suggested to be one of the key criteria

12:57.405 --> 12:59.065
of a, of a course of relationship.

12:59.605 --> 13:01.065
Uh, it does that really well.

13:01.245 --> 13:04.305
Uh, but for its bite, it also has its own limitations

13:04.305 --> 13:08.505
because it is, it gives you those regularly,

13:08.575 --> 13:09.745
occurring connections,

13:09.885 --> 13:12.225
but then again, it, it struggles to deal

13:12.225 --> 13:13.425
with cause of dependence.

13:13.425 --> 13:16.345
There may be spurious correlations in there.

13:16.925 --> 13:18.065
And, and this is again,

13:18.065 --> 13:21.665
where the other approaches in combination can compliment

13:21.665 --> 13:23.745
and offset some of the limitations,

13:23.745 --> 13:24.865
that each of them have.

13:26.075 --> 13:29.135
Um, I think the, the future, so the,

13:29.135 --> 13:32.255
the future implication of this is really around,

13:32.255 --> 13:35.975
methodological practice, so that as social scientists

13:35.975 --> 13:38.295
and as management researchers,

13:38.435 --> 13:41.295
we should really embrace this idea that if we want

13:41.295 --> 13:42.295
to make robust calls

13:42.295 --> 13:45.935
or in infants, there's not one single route is not

13:45.935 --> 13:48.495
what single approach to cause and identification,

13:48.555 --> 13:52.535
but we really benefit when we, in our,

13:52.645 --> 13:55.135
inquiries into particular phenomena make the most

13:55.135 --> 13:56.735
of these different approaches together

13:56.955 --> 14:00.015
and use them alongside each other in ways

14:00.015 --> 14:03.615
that offset the biases and limitations of the one approach,

14:03.715 --> 14:08.015
but also in, in simultaneously complement each other in,

14:08.035 --> 14:10.935
in moving towards these more robust causal in infants.

14:11.655 --> 14:14.955
Um, so that is the idea with causal triangulation, where

14:15.635 --> 14:17.875
previously we talked a lot about causal identification,

14:18.055 --> 14:21.225
and that's an approach that primarily comes out

14:21.225 --> 14:22.265
of econometrics

14:22.445 --> 14:24.465
and is a set of techniques that fit

14:24.465 --> 14:27.545
with the propositional approach to theorizing,

14:27.545 --> 14:30.545
that has its, its its value and its advantages,

14:30.545 --> 14:33.985
but as I've explained, is still, limited

14:34.675 --> 14:36.465
given the way in which it tends

14:36.465 --> 14:39.305
to think about causal relationships in a very narrow and,

14:39.305 --> 14:40.545
and myopic manner.

14:41.165 --> 14:45.775
Uh, so to help with that, to compliment that approach,

14:46.275 --> 14:49.655
and to, to expand it, I suggest that we should,

14:49.655 --> 14:50.695
should still

14:50.715 --> 14:51.765
Use those techniques,

14:51.865 --> 14:54.965
but use them alongside the techniques that we have,

14:54.965 --> 14:56.965
with the other two approaches at the same time.

14:57.545 --> 14:59.965
And if we do this well as social scientists

14:59.965 --> 15:01.885
and as management researchers, so that we,

15:01.985 --> 15:04.405
in a programmatic manner, systematically

15:05.225 --> 15:08.645
use them alongside one another for particular topic areas,

15:08.655 --> 15:12.045
particular phenomena that we're studying, I have no doubt

15:12.045 --> 15:14.925
that will come up with stronger causal in infants, that

15:15.035 --> 15:17.125
that will lead to better explanations,

15:17.265 --> 15:20.005
and as I said, will also inform policy and practice.

Understanding Causal Relationships in Management Research

00:00:00 - Causal relationships, have been a concern within management research and across the social sciences for, for a long time. and of causal relationship is really the idea that for the events are happening in the world that we study, we can really say that one, condition A is seen to produce a particular effect, b and that seemed to be important to be able to establish those causal relationships so that in no uncertain terms and beyond basic correlations, we can say a will on
00:00:33 - either in a very direct deterministic sense that A, always B always follows from a, the one condition, or the one effect from, the base condition, or maybe in a more probabilistic sense that a on average is more likely to, to lead to B. So this has been a ma a main preoccupation of, many social scientists, including management research for a very long time. Also, with the perspective that if we, if we're able
00:01:04 - we can also more effectively inform policy and practice on the things that, managers or policy makers might, might do. Yeah. What is interesting is that the topic of causation, and also questions around causal relationships, and also the idea that we need to be really certain, be able to make a robust inference as to whether a causal relationship is there. These types of questions haven't really been su substantively discussed, to any great extent.
00:01:36 - that has taken place is really at the level of methods where, typically by importing, techniques and methods and research designs from econometrics or from sociology, people have made assumptions about what causal relationships are and how they can following these methods, be properly identified in each of these different contexts. But what is important to remember here is that, if you compare, let's say, a experimental design with a cross case comparative analysis method,
00:02:11 - what a causal relationship is, and the criteria that are also coming with these methods as to how to make a causal inference, how to make an inference that a causal relationship is there, are actually very different. and that discussion hasn't really taken place yet. And I think it's important to first figure that out how we generally think about causal relationships, alongside applying some of these methodological techniques. yeah, that's, it is a very good question.
00:02:43 - that there are three ways in which we tend to approach questions of causation and tend to approach the, the issue of how to identify cause relationships. So there's a propositional approach, which is basically what it says where researchers Hypothesized that, two events or, happenings are related as a simple cause effect scenario. So one condition, A, directly effects B, either in a direct way or through some mediating path.

Exploring Causal Approaches in Research Design

00:03:16 - So you have these very simple if done scenarios where researchers subsequently do their best through their research designs and through their, intervention techniques to make sure that in a very controlled manner, they can actually establish that b in no uncertain terms follows from, from a, so that's the propositional approach. Then we have the configurational approach, which actually emerged as, an antidote
00:03:45 - where the propositional one is really looking at a root cause or root condition, or presupposes a root cause or root condition, and then looks at the subsequent effect In a very simple sort of causal scenario, the configurational approach says, well, actually, what might be better given that many phenomena are very complex, is to look at a particular outcome or effect that has materialized and then look backwards at all the sets of conditions
00:04:17 - that might have brought that particular effect about. So it's a very different approach where instead of looking at the effects of presupposed causes, you look at what might be the potential causes or the potential conditions that might have caused, a particular effect. And it's as an approach also, compared to the propositional one, much more alive to a complex causal field, a whole range of conditions
00:04:45 - that might have brought the effect about. So your frame of reference, is enlarged, but you also with that look at multiple conditions alongside one another, and you try and figure out through your reasoning and through your probing of what you found. so typically these, this is post case comparisons that you, or, or comparative research design that you use across cases or across data sets where you try and not just identify these potential conditions,
00:05:20 - or which of the sets of conditions alone or together might have brought this effect about. So it's a much more systemic form of causation compared to the, the propositional approach. And then finally, the process approach, is, is again, another approach. So it says, instead of listing sets of conditions on the one hand and effects on the other, either in a very simple way with the propositional approach or in a more complex way with a configurational approach,
00:05:52 - to follow in a more longitudinal sense how hap how events play out over time so that we can really see what is the sequencing of events that might have materialized in a particular case or cross cases, and maybe also across larger data sets, if we have that longitudinal, frame of reference for those data to see what effect materialized in that particular case. So it's a, an approach that takes temporality

Understanding Causation: Different Methodological Approaches

00:06:25 - and is also alive to the very uncertain probabilistic nature of the world where sometimes effects get brought about through certain sequences of events that, that happen. So they have their own, that approach comes with its own methodology, it's longitudinal, but there's also various techniques that you can use to, to bracket sequences of events to see the transition points between these, these sequences and to try
00:06:56 - alongside that trajectory. Yeah, I think, I think the most interesting thing is that, so each of these approach approaches obviously comes with its own, prepositions of how to identify causal relationships. And it also makes very different assumptions about where we find these causal relationships and where and how causation happens. And one, obvious observation is, and there's a philosopher called Nancy Cartwright, who
00:07:28 - that some approaches are really looking at, so she calls that the really, really, using causes or manipulating causes. So think of the propositional approach where you say, okay, I'm assuming this is a, a root cause or a true cause that will bring a certain effect about either in a deterministic or in a more probabilistic sense. Then I'm trying to find research designs and data sets that allow me to manipulate
00:08:01 - And then, so it's really the look looking at the effects of presumed cause then the other approaches, so the process and configurational are actually the opposite, so it's more hunting for causes. So I know that these types of effects as phenomena generally occur in the world. Let's look back and see what might have been, all the sets of conditions and, and potential causes, in those terms
00:08:30 - So it's more hunting for causes, trying to identify and, and weed out the sets of conditions, that may have, may have not mattered too much. Then the sets of conditions that we now can say are actual causes that produce this, this outcome. So it's, it's as an approach compared to the using or manipulating causes and looking at these very controlled intervention, experimental based designs that match that.
00:09:04 - and looking more at the world out there, so phenomena that can be observed effects or outcomes that can be observed and then be reasoning backwards. So I think that's a very fundamental distinction, and it's actually very helpful to think about it because it also allows you to say where the one comes with a blind spot. You know, if I only reason forward and presume that condition A will BA root cause and will in most instances lead to B, well,

Integrating Approaches for Robust Causal Inference

00:09:34 - and we have something to learn from the other approach. not to say that, you know, I'm not trying to favor the one over the other, but, the point with the triangulation idea is to say that they all come with different vantage points. They all come with different strengths and limitations and also different criteria for making, an influence around the causal relationship. So let's join them up, let's see what they collectively offer so that we get a more robust influence as
00:10:05 - accordingly theorize better come up with stronger explanations, but also inform policy and practice in a, in a much more effective way. and to back that up, in the paper, I show how some of the classic criteria for how we think about, for how we think about a cause of relationship, so what a cause of relationship needs to have in order for us to talk about not just the correlation, but a course of relationship, how these criteria are not
00:10:37 - but they're more likely to be satisfied by multiple approaches. So the propositional one is very good with its controlled intervention methodology to try and identify whether a course temporally precedes an effect. So the course or precedence that is a classic criteria of, of acausal relationship is, is something that that approach can do really well, but it also has a lot of limitations. I already noted the lack of its ability to deal
00:11:10 - with spurious causation. The lack of its ability to really beyond doubt show that there is a dependent relationship between A and B. So B will necessary and sufficiently follow from, from a, those are all things that it struggles with. Whereas the other approaches then have, have equity strengths and limitations. So the process one is very good at, as I described it, to trace the, the real transitive causal process that goes
00:11:42 - or the events that you've, you've categorized as conditions or potential causes over time. So tracing that through. So the classic notion that we can speak of causation when A, produce an, an effect, and B, b produce an effect C, you can trace that or be much closer to that with the process approach than, with the other two approaches. But equally, the process approach also has its limitations, and again, causal dependence to really showing that b
00:12:16 - and of itself by that approach. And then the configurational approach is very good in trying to force yourself to look at the world out there, not limit yourself to one presupposed condition or limited set of conditions as potential causes, but really look at the world out there. What are the regularly occurring connections between potential causes, conditions as causes, and the effects that that materialize.

Enhancing Causal Understanding Through Methodological Triangulation

00:12:49 - in the events that we observe and the connections between them, which is, which has for a long time been suggested to be one of the key criteria of a, of a course of relationship. it does that really well. but for its bite, it also has its own limitations because it is, it gives you those regularly, occurring connections, but then again, it, it struggles to deal with cause of dependence. There may be spurious correlations in there.
00:13:18 - where the other approaches in combination can compliment and offset some of the limitations, that each of them have. I think the, the future, so the, the future implication of this is really around, methodological practice, so that as social scientists and as management researchers, we should really embrace this idea that if we want to make robust calls or in infants, there's not one single route is not what single approach to cause and identification,
00:13:52 - inquiries into particular phenomena make the most of these different approaches together and use them alongside each other in ways that offset the biases and limitations of the one approach, but also in, in simultaneously complement each other in, in moving towards these more robust causal in infants. so that is the idea with causal triangulation, where previously we talked a lot about causal identification, and that's an approach that primarily comes out
00:14:22 - and is a set of techniques that fit with the propositional approach to theorizing, that has its, its its value and its advantages, but as I've explained, is still, limited given the way in which it tends to think about causal relationships in a very narrow and, and myopic manner. so to help with that, to compliment that approach, and to, to expand it, I suggest that we should, should still Use those techniques,
00:15:17 - with the other two approaches at the same time. And if we do this well as social scientists and as management researchers, so that we, in a programmatic manner, systematically use them alongside one another for particular topic areas, particular phenomena that we're studying, I have no doubt that will come up with stronger causal in infants, that that will lead to better explanations, and as I said, will also inform policy and practice.

Faculti AI can make mistakes

Research

Policy

Analysis