[roenxi]

1. I don't think that is a meta-study. It seems to be an attempt to build a dataset to track US union membership over long timeframes.

2. It notes that there is a correlation between union membership and inequality. Which is interesting but not that powerful - correlation is not causation. It might be that both trends are being driven by the financialisation of the US economy.

3. It finds that union households earn a premium over non-union households. Again, because of the nature of the study that doesn't tell us much about the impacts of unions. As an analogy, we might find that HN commentators earn more than non-HN commentators in the tech industry but that doesn't indicate that HN is pushing salaries up.

Although in fairness I would suspect there probably is a causal element. But I still don't want to be in a unionised industry. I don't want a premium over other tech workers. I want to maximise the average tech worker salary and then be employed in tech. Those are very different objectives and require different strategies to achieve.

Pro-union types tend to have a very short term view of the world and aren't about maximising long term returns. Strikes and collective bargaining don't move the needle in the right direction over the long term.

[aesh2Xa1]

You're right that this isn't a meta-study. It's a deep survey and references many similar studies, though. It certainly doesn't exist in a vacuum.

"In this section, we explore in a more direct manner the relationship between unions and income inequality, joining an extensive empirical literature examining how unions shape the income distribution."

You're right to point out that "correlation is not causation," but the study specifically addresses your concern and presents a strong argument for causation. It's not as if science can never demonstrate anything using correlation and statistical techniques, you know?

https://xkcd.com/552/ "Correlation doesn’t imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing ‘look over there.'"

> It notes that there is a correlation between union membership and inequality

Specifically, it notes a robust INVERSE correlation that:

* Increased union membership correlates to decreased inequality

* Decreased union membership correlates to increased inequality

> Which is interesting but not that powerful - correlation is not causation

The authors acknowledge the statistical nature of correlation, and they addressed it (they say so right in the abstract). They used the following techniques to establish a causal relationship between union density and income inequality.

* distributional decompositions

* time series regressions

* state-year regressions

* instrumental-variable strategy based on historical events like the 1935 legalization of unions and the World War II–era War Labor Board

> It might be that both trends are being driven by the financialisation of the US economy.

The authors find that policy changes which significantly reduced the cost of union organizing (e.g., the Wagner Act and the National War Labor Board during WWII) led to lasting increases in state-level union density and corresponding reductions in income inequality. These effects were specific to the periods when these policies were active and had no similar impact in other times, such as during the Korean War, which did not explicitly promote union organization. This is a cause very different from a wild guess at "financialisation of the US economy."

Furthermore, the study highlights that unions were particularly effective in reducing inequality by increasing the wages of less-educated and nonwhite workers. During periods of high union density, the wage gap between union and nonunion workers was substantial, contributing significantly to overall income equality.

> It finds that union households earn a premium over non-union households. Again, because of the nature of the study that doesn't tell us much about the impacts of unions.

The study does more than just observe a premium; it provides historical and statistical context to argue that the premium is associated with union activities, such as collective bargaining. The consistent premium over many decades, despite changes in union density, suggests a link between union presence and wage levels.

In short, no, neither union membership or inequality are evidenced as show "both caused by financialization of the economy." They the latter correlated to the former, and was tested for causation. The former's uptake can include economic considerations, but it also correlated directly to governance policy directly targeting labor law.

[roenxi]

> They used the following techniques to establish a causal relationship between union density and income inequality.

I don't think those techniques establish causal relationships. Which of these techniques do you think establishes a causal relationship? I can tell you right from the start that "time series regressions" don't establish a causal relationship. The paper established a very strong statistical relationship.

Which is all very well but if you look in the paper [0] at Fig I you can see a very strong statistical relationship without any need for statistical methods. It leaps out of the graph at you. There was a pre-WWII period, the WWII-through-to-US-mini-peak-oil in the 70s and then the post-peak [1] regime (speaking loosely since shale oil has indeed been a miracle over the last decade - but it isn't the wealth engine that oil was back post WWII). There are a lot of interesting statistical correlations at around the same time.

That is far too much background noise to claim that unions are the causal element. Geopolitics and cheap energy was more likely to causal.

[0] https://www.nber.org/system/files/working_papers/w24587/w245...

[1] https://wtfhappenedin1971.com/

[aesh2Xa1]

> Which of these techniques do you think establishes a causal relationship? I can tell you right from the start that "time series regressions" don't establish a causal relationship.

Hm, time series series regression is a standard, accepted approach to causal inference:

https://towardsdatascience.com/inferring-causality-in-time-s...

For me, it suffices to say that the authors did not weakly position their argument as you claimed. I responded because I thought that claim was an attack, and that it was a careless regurgitation of the standard line about correlation.

There's some author discussion here that might help get the points across:

https://www.stone-econ.org/research/unions-and-inequality-ov...

Here, also, is a third-party discussion:

https://journalistsresource.org/economics/inequality-labor-u...

The data that the union papers authors used is here. Scroll to the bottom of the page to find

"Supplementary data | qjab012_Online_Appendix - pdf file"

https://academic.oup.com/qje/article/136/3/1325/6219103

[roenxi]

> Hm, time series series regression is a standard, accepted approach to causal inference

But statistical causality - things like Granger causality for example - aren't, in reality, establishing causality. They're statistical properties. You can't ever establish causality from statistical data. Eg, if I light a log on fire there will be bright light and later on there will be ash. If you have a timeseries of luminosity and quantity of ash present, bright light will be Granger-causal of ash. But in reality we know that bright light isn't causing the ash; the situation is we are analysing a bonfire.

You've got a group of people there in that analysis article that aren't very good at interpreting results. They're looking at a time of extreme turmoil, they've picked 2 random timeseries that are responding to underlying causes and assuming that they are the entire story. They can't do that, it isn't a valid argument. It isn't a thorough enough treatment. In analogy, they're missing the fire for the light. There isn't particularly strong evidence that unions do anything on their own at the macro level; especially since the economic regime was just very different in an era where the available energy supplied was cheap and quantity was rapidly increasing.

> For me, it suffices to say that the authors did not weakly position their argument as you claimed.

I never said they weakly positioned their argument, their argument is watertight, they developed a data set and analysed it. Found a bunch of interesting statistical facts. Solid academic work. camdat weakly positioned his argument.

[aesh2Xa1]

The authors are using other historical events to help improve the theory. They aren't solely reliant even on time series.

This isn't an experimental study, and so they have to rely upon plausibility in context. This explains their multi-faceted approach a la distributional decompositions and state and IV.

To me, the contrarian position — that unions have no such effect — doesn't look as good. Prove it :)

[roenxi]

They don't. I think there might be a gap between what they wrote and what you think they wrote. They aren't attempting to rely on "plausibility in context", they're doing academic work and they're stating basic facts - they developed a dataset and analysed it. That analysis revealed a bunch of interesting statistical features. But that is a series of fairly specific statements. What they aren't claiming is to have a theory. There isn't a theory in the paper. They aren't doing any work that requires theorising. They're just looking for evidence.

And they found some, but it is weak evidence for the idea that unions have a positive influence and it is unclear what it actually shows in reality. It is a good example of the truism that correlations are not causations.

> To me, the contrarian position — that unions have no such effect — doesn't look as good. Prove it :)

I do believe that unions have a generally negative effect, but that isn't what I'm arguing about in this thread. My point here is that this paper isn't a meta study and is evidence of something different than what camdat originally claimed. And I felt your response was interesting enough to justify a few extra comments about the difference between statistical causality and practical causality.