Interpreting the Systemistas-Randomistas debate on development strategy

Note: You may notice that this post is dated older than when this website was created. That is because, this is a cross post. I had originally posted this on the Effective Altruism forum here where I received comments disagreeing with some of the points I make on this post. I am cross-posting it here in the spirit of collecting everything I have written in one place.


There has been a debate on development strategy emerging between development economists[1][2]. While one portion of development economists perform RCTs (Radomistas) another portion work on systemic changes (Systemistas, as I would like to label them) and the ongoing debate between them is on which strategy is ‘better’. There has already been a post on this forum[3] which presents the arguments from the Systemistas perspective. Below I provide my interpretation of the thread of arguments in this debate without necessarily taking sides. A summary of my interpretation is that Systemistas and Randomistas take different approaches to build their models of economic growth which takes them down very different paths. If you have already heard the arguments in this debate before, one new thing you might learn from this piece is the growth model that the Randomistas might have in mind.

This debate is relevant to EA since the community spends a lot of resources on RCTs and views that are for or against it can be important. In addition, the direction of this debate could also have implications on how we think about human well-being at large.

Epistemic status

I am not an economist (atleast yet). Clarity over correctness has been the order of priorities here. Since it is my interpretation of the debate, in what follows, whenever I write to the effect of “Systemistas say..” or “Randomistas say..”, you must take it to mean “The author believes Systemistas say…” or “The author believes Randomistas say…”.

The problem with growth models

Development economists who are Systemistas believe that economic growth is the best way to improve human well-being. Their passion on the subject is best expressed by the following quote from [4]:

“Within the advanced countries, growth rates tend to be very stable over long periods of time … For poorer countries, … there are many examples of sudden, large changes in growth rates, both up and down. … I do not see how one can look at figures like these without seeing them as representing possibilities. Is there some action a government of India could take that would lead the Indian economy to grow like Indonesia’s or Egypt’s? If so, what, exactly? If not, what is it about the ’nature of India’ that makes it so? The consequences for human welfare involved in questions like these are simply staggering: Once one starts to think about them, it is hard to think about anything else.”

By economic growth, Systemistas typically mean growth in GDP per capita. Changes in GDP per capita correlate positively with a lot of indicators of human well-being [3]. Of course, one can think of more holistic measures of economic growth other than GDP per capita. But I find it hard to imagine how any such attempts could negate the idea of economic growth being positively correlated with human well-being.

But correlation and causation are different things altogether. Is it that economic growth creates human well-being or human well-being creates economic growth? Both these directions of causations can explain the correlations. Further does the causation run the same way in all countries? And what about the level of causation in different countries? Growth economists attempt to answer such questions by building models of economic growth and checking it against empirical data. The latter is often done by performing regressions of all sorts with macrolevel cross-country datasets. There is a mountain of literature on this topic. But the simplest model is the Cobb-Douglas form of the aggregate production function below. This is also the basis of ‘Growth accounting’:

$$ Y = A \times K^\alpha \times L^{1-\alpha} \tag{1} $$ where $Y$ is the total production (basically GDP at the aggregated level, say, of a country), $A$ is the Total Factor Productivity (hitherto, TFP), $K$ is capital accumulated, $L$ is raw labor available and $\alpha$ is the capital’s share of contribution to the output. Equation (1) is just one specific form of the aggregate production function. A general aggregate production function is of the form: $\mathcal{F}\left( A, K, L\right)$. In the literature they put the capital and labor terms together and call it “Factor Accumulation”. All of this is to say, that economic growth could be accounted for by two things: Factor (capital of all sorts & labor) Accumulation and Total Factor Productivity.

The notorious TFP

The next natural step is to ask: Which among these two things - Factor Accumulation or Total Factor Productivity - explains growth better? There is a definite answer to this - “Factor accumulation does not account for the bulk of cross country differences in the level or growth rate of GDP per capital; something else- TFP - does” [5] So TFP explains growth, atleast in the long run. But what is TFP and how can we explain the differences in TFP across countries?

This is where doubts creep in and the skepticism of the Randomistas begin. It is not entirely clear from the literature what explains “the TFP blackbox”[6]. You can even see this confusion expressed in the aforementioned quote which refers to TFP as ‘something else’. Different economists attribute different meanings to it. It has been interpreted as Technology, externalities of human capital, coordination failures (For a discussion on these three interpretations see section 3 of [7]) and so on. It is not entirely clear to me what is the general level of consensus on each of these interpretations.

In addition, the model also predicts something called Convergence. That is, poorer countries must grow faster and eventually catch-up with the richer countries. But, if you don’t live under a rock, you already know that this is not happening, atleast not in all the poorer countries. What we see instead is “Divergence, big time”[8]. Valiant attempts have been made to explain this divergence[9], in vain[10], and attempts continue to be made.

Such theoretical problems make it hard for the Systemistas to give precise prescriptions on what needs to be done at a macro level. This lack of a precise prescription is possibly what makes the Randomistas highly skeptical of the Systemistas. The failure of prescriptions like the Washington consensus[11] add to their skepticism.

Skepticism v Cynicism

But being skeptical about growth can’t justify throwing away the macro idea of growth-leading-to-welfare altogether and focusing on micro interventions through RCTs. This dialogue between Shruti Rajagopalan and Lant Pritchett in a recent podcast[12] makes this point:

PRITCHETT: … One thing I’m very much worried about is that the backlash against economists perhaps being overly dogmatic about there being a single, narrow recipe for growth has [not] been … “We need a different recipe,” but “We don’t need growth.” It’s like if you say, “In order for you to have a healthy heart, you need to eat the following diet.” It could well be that people are just wrong about how narrow the necessity of that diet is, in which case you could have some skepticism. If the response to that was, “I don’t like that diet; therefore, I’m going to weigh 300 pounds and never exercise.” That’s really bad for your health.

RAJAGOPALAN: Skepticism about the lack of the poor diet is not the same thing as skepticism about the need for a healthy heart, right?

PRITCHETT: Exactly. I feel there’s way too much skepticism about the need for growth, and I think that’s just a deep confusion. …

(Emphasis added by me)

While it is true that growth models are far from perfect, the right response to that might not be to stop making growth models altogether. It might just be that we need to figure out different techniques to build better models. A Systemista could argue that given the stakes that are facing us here, maximum resources must be spent in doing this instead of doing RCTs.

The Randomistas’ growth model

A perfunctory reading of the arguments does make it seem like the Randomistas are throwing away the growth-leading-to-welfare idea. But as I see it, the Randomistas have a growth model. They use a different approach to building it and that leads them to look at microdata and hence RCTs. Below I will attempt to give just a flavor of their thought process. If you are ’non-mathy’ feel free to skip these equations (including the paragraph that explains the symbols in it) and read what is below it.

These equations are taken from [7] with some changes in notation intending to put human capital and physical capital inside one term $K$ (and not inside the $A$), for simplicity’s sake:

$$ \text{The aggregate production function at the macrolevel,} \mathcal{F}\left( A, \overline{K}, \overline{L}\right) \tag{2} $$

$$ \text{The production function at the microlevel, } F\left(\theta, K, L\right) \tag{3} $$

$$ \text{The distribution of productivity across the population is, } \tilde{G}(\theta) \tag{4} $$

$$ \text{Now, } \mathcal{F} \left( A, \overline{K}, \overline{L} \right) \equiv \max_{K(\theta), L(\theta)} { \int_{\theta} F(K(\theta), L(\theta), \theta) d\tilde{G}(\theta) } \tag{5} $$

$$ \text{subject to, } \int_{\theta} K(\theta)d\theta = \overline{K} \text{ and } \int_{\theta}L(\theta)d\theta = \overline{L} $$

It is important to see that equation (2) is of the same generalized form of the aggregate production function (with bars on top of $K$ and $L$ to indicate aggregation) which I had mentioned earlier when discussing equation (1). What is new here is equation (3) which is the production function of every micro level individual. In case of individuals ‘TFP’ is interpreted as productivity of each individual. It is represented as $\theta$. Equation (5) creates an equivalence between the macrolevel production function in equation (2) and the microlevel production function in equation (3) by maximizing the integral of the latter. This integral is done with $\tilde{G}(\theta)$ (from equation (4)) and by doing so we are basically disregarding the individual productivities which is actually an implicit assumption in the aggregate production function that didn’t come out clearly in the earlier discussion of equation (1). The final line just tells us the constraints of the optimization. Overall, from a math perspective, nothing much has really changed in this version: the generalized form of the aggregate production function is still $\mathcal{F}\left( A, K, L\right)$ (with a couple of bars for notational convenience). But we are now looking at it as the sum of micro level production functions.

Here is the interpretation of these equations. The idea is to imagine every individual in the economy (it can be people, firms et cetera) as having their own outputs and hence their own capital and ‘TFP’ (which has a clear interpretation for an individual as their productivity). From the Randomistas perspective, the economy’s output as a whole must be a sum of these individual outputs. So the capital of the economy is a sum of the individual capitals and the total labor in the economy is a sum of the individual labor.

Once you accept this premise, you start seeing a lot of the assumptions that were implicit in the macrolevel growth model starting to pop up. One of the big implicit assumptions is that the macro models don’t necessarily care who has the endowments within the country to start firms. They assume markets are perfect (see first welfare theorem) and that the person with the right level of productivity will be able to trade for the right amount of capital which will create the right kind of firms which will then create the growth needed for convergence. But the Randomistas, who work a lot with micro data, know this to not be true. Market/government failures are very real. A lot of empirical work has been done to show this to be the case. This leads to the growth theory of the Randomistas: Macrolevel growth must be the sum of microlevel growth.

But the Randomistas growth theory is far from complete. Even if one completes it, we do not possess the rich microlevel data that might be needed to empirically verify such models. So, the Randomistas say, let us collect this data now by designing experiments using RCTs and we will figure out how to build the model with it later.

Public Policy implications

The Systemistas counter by saying that RCTs are not methodologically sound. They claim it won’t be possible to generalize RCT results[13] which would then make it difficult for building any growth model that the Randomistas may have in mind. Further the Systemistas see issues when it comes to public-policymaking. The Randomistas say RCTs provide precise evidence to policymakers. But it is easy to make bad policy even with precise evidence. To illustrate this point, let us consider a policy solution has been proposed - the government shall distribute $n$ kg of free rice to the poor nationwide. Imagine a policymaker who is a Randomista. This policymaker would jump right into designing-an-RCT mode since it will give precise evidence on the logistics of this rice distribution program. But is that what a policymaker should do? Should not the policymaker first ask questions like, “Is it right for the government to be intervening here and providing rice?”? Since an RCT can’t be run to answer such a question the Randomista policymaker might skip this altogether and run the program. But if it turns out that this was a case of inappropriate government intervention then the policy would still fail at a macro level even though the logistics of it were designed using precise evidence from an RCT.


So that is my interpretation of the debate. The Systemistas approach to growth modelling is ‘very’ top-down while the Randomistas take a ‘very’ bottom-up approach.

If you have read this far you are probably someone interested in some prescriptions! Below I give some. If you agree with my interpretation (which if you don’t, then we end up with a fun meta-debate!):

  • and you are a Randomista, then it will be useful if you prove that macrolevel models can indeed be built using experimental results and that such models can provide prescriptions leading to sustainable human well-being.
  • and you are a Systemista, then it will be useful to prove that Randomistas’ growth models just won’t lead to sustainable human well-being.
  • and you are a policymaker, then don’t jump the gun! Remember that precise evidence is important but it is not everything.
  • and you are an interested third party like me, maybe you can go meta with a Scientometric analysis on the literature here and observe the trends of this debate. You can see if resources (number of papers, number of researchers etc.) are skewed towards one side of this debate and look at the economics of the debate play out!

Overall, I believe, the arc of development economics is long but it shall eventually bend towards sustained human well-being. Meanwhile, RCTs shall be run and cross-country regressions shall be done. The debate will continue to play out.


  1. Deaton v Banerjee. NYU Development Research Institute. Retrieved March 29, 2022, from
  2. Pritchett, L. (2017, March 28). Getting Kinky with Chickens. Center for Global Development.
  3. Hillebrandt, H.; Halstead, J. G. (2020, January 16). Growth and the case against randomista development. Effective Altruism Forum.
  4. Lucas, R. E. (1988). On the mechanics of economic development. Journal of Monetary Economics, 22(1), 3–42.
  5. Easterly, W.; Levine, R. (2001). It’s Not Factor Accumulation: Stylized Facts and Growth Models. The World Bank Economic Review, 15(2), 177–219.
  6. Banerjee, A.V., 2008. Big answers for big questions: the presumption of growth policy. In Brookings Conference What Works in Development.
  7. Banerjee, A. V., Duflo, E. (2005). Growth Theory through the Lens of Development Economics. Handbook of Economic Growth, 1(SUPPL. PART A), 473–552.
  8. Pritchett, L. (1997). Divergence, Big Time. Journal of Economic Perspectives, 11(3), 3–17.
  9. Mankiw Gregory, N., Romer, D., Weil, D. N. (1992). A Contribution to the Empirics of Economic Growth. The Quarterly Journal of Economics, 107(2), 407–437.
  10. Klenow, P. J., Rodríguez-Clare, A., Rodriguez-Clare, A. (1997). The Neoclassical Revival in Growth Economics: Has It Gone Too Far?. NBER Macroeconomics Annual, 12, 73.
  11. Rodrik, D. (2006). Goodbye Washington Consensus, Hello Washington Confusion? A Review of the World Bank Economic Growth in the 1990s: Learning from a Decade of Reform. Journal of Economic Literature, 44(4), 973–987.
  12. Rajagopalan, S. (2022, March 17). Ideas of India: Where Did Development Economics Go Wrong?. Discourse.
  13. Pritchett, L. (2021). Let’s Take the Con Out of Randomized Control Trials in Development: The Puzzles and Paradoxes of External Validity, Empirically Illustrated (No. 399; CID Faculty Working Paper).

Other sources

I skimmed a lot of stuff but I couldn’t cite all of them. So just wanted to add a couple more. If there are other resources you know, please put them in the comments too:

  1. Banerjee-Duflo’s 2018 course at the Paris School of Economics found here. The Banerjee lectures in this course helped me read parts of [7]
  2. Debraj Ray’s entry in the New Palgrave Dictionary of Economics and his textbook on development economics explain the evolution of growth theory quite well. If you thought this article was interesting (or not), you would just love his writing!