Myth of productive failure

 

Yesterday I watched an excellent video by Matthew Burns. LINK

Extracts from the video:

When people learn, they go through four very specific phases.

We start off in the acquisition phase. We're slow and inaccurate. No idea what to do. So, when people are in the acquisition phase, the best thing we can do is do immediate modeling with explicit instruction.

The task there is to get them accurate. Once we get them accurate, then we can work on speed, proficiency. 

Once you're able to do it with sufficient speed and accuracy, then they can generalize it to some other setting. What matters is automaticity.

You can't generalize something unless you can do it with automaticity. How do we know if it's automatic? Speed. That's why speed matters. What really matters is automaticity. But we have to get accurate first, then sufficient speed then you can generalize the skill and then once you can generalize it you can use the information to solve problems.

This is the basic learning theory. The basic steps in human learning for kids through adults through children learning their basic letter sounds to 16-year-olds learning how to drive a car. We, generally speaking, go through these phases and we want lots of practice but only after we reach some level of accuracy.

 

My thoughts:

This aligns precisely with what I've been advocating in my blog posts since 2010. As early as 1910, Edward Thorndike emphasized the paramount importance of initial input in shaping learning outcomes, drawing on principles like the law of effect and readiness to underscore how early experiences form the foundation for all subsequent acquisition.

This idea is vividly echoed in Charlie Munger's analogy: "The human mind works a lot like the human egg. When one sperm gets into a human egg, there's an automatic shut-off device that bars any other sperm from entering."

In educational terms, once a child internalizes an incorrect concept—say, a muddled letter sound—it creates a cognitive barrier, making it far harder for accurate information to penetrate later. This reinforces why explicit, modeled instruction must precede any unstructured exploration in early learning stages. 

 

Extracts from the video:

Now how does that fit in the conversation about productive struggle?

This is productive problem solving followed by instruction. So that means give them the task, let them try and solve it first, experience their productive failure. And that's called problem solving followed by instruction as opposed to follow instruction then problem solving.

But most importantly, look at the effect for second to fifth graders. (You can get this information from the video above) Negative.09. Negative. 09 does more harm than good. The instruction, then problem solving did better than the problem solving then instruction. Productive failure was not effective.

It is inaccurate to look at these data and say a productive struggle approach was effective for students in second through fifth grade. It wasn't even reliably effective for sixth through 10th graders. It was better, but not reliably so.

My thoughts:

The negative effect size of -0.09 for second- through fifth-graders in productive failure approaches—indicating not just ineffectiveness but actual harm compared to instruction-first methods—is stark evidence against applying this strategy in early elementary years.

This is particularly evident in reading instruction, where a significant portion of children—around 60%—don't achieve proficient reading until between fourth and sixth grades, often after years of frustration that leads to shut down.

Such delays aren't benign; they correlate with higher rates of disengagement, where kids internalize failure as a personal deficit rather than a temporary gap. Introducing "productive" struggle too soon exacerbates this, turning potential learners into avoiders before they've built the foundational automaticity needed for deeper problem-solving.

 

Extracts from the video:

I was one time recently observing a kindergarten classroom and the kindergarten teacher was teaching the letter T. Kids have never seen the letter T before and she's a good teacher using a good instructional program. And she stands in front of her students and says "Okay kids, today we're going to learn something new. We're going to learn what this is." She puts a Letter ‘T’ on the Smartboard. "Okay, who knows what this is." They start getting all squirmy. Okay. This is the letter T. Who knows what sound the letter T says? They didn't know it was a letter T. Are they going to know what it says?" No.

Now, I think she was trying to operate this idea of productive struggle. Let the kids try and figure it out. But let's go back to learning theory. Productive struggle, I would argue, works more so, if at all, down here (I think he means the attendees). When someone is first learning something, modeling explicit instruction, then lots of practice, then opportunities to generalize it, followed by using it to solve problems. Once the kids demonstrate good initial learning and they can generalize it, then you want to have them engaged in productive struggle. I think that's probably okay. I haven't studied it, but I think that's probably okay. I would argue productive struggle has no place in initial learning. All of that way before we ever consider something along the lines of productive struggle. I would have done this.

My thoughts:

Absolutely—productive struggle has no place in initial learning, especially at the primary school level. For novices, particularly young children, this approach risks overwhelming them with confusion rather than sparking insight. Research on learning phases confirms that acquisition demands direct modeling to build accuracy before any fluency-building practice, let alone generalization or application.

Introducing struggle prematurely can entrench misconceptions, aligning with Thorndike's warnings about the fragility of early neural pathways in education.

Instead, reserve it for later stages, once automaticity is secured, to avoid the cognitive overload that leads to disengagement.

 

Extract from the video:

I would say, "Okay, kids, today we're going to learn a new lesson. A new letter. It's a letter. Put the letter T up. This is the letter T." Everyone, say it with me. T. They'd say T. Good. T. T makes the /t/ sound. Okay, say it with me. T makes the what sound? T makes the /t/ sound.

My thoughts:

This scripted explicit instruction is extremely important. Please listen to minutes 8 and 9 of the video above. Burns' crisp pronunciation of letter sounds—isolated and pure, like /t/ without blending into extraneous sounds—stands in stark contrast to common errors highlighted by Mark Seidenberg in works like Language at the Speed of Sight. If all teachers adopted this precise phonemic modeling, we could drastically reduce the incidence of "shut-down" kids who disengage due to persistent decoding confusion. Seidenberg's analysis shows how blended or elongated sounds (e.g., saying "tuh" instead of pure /t/) distort phonemic awareness from the outset, compounding reading delays.

Prioritizing this fidelity in initial modeling isn't pedantic—it's preventive medicine for literacy.

 

Extract from the video:

At 9:08 minutes T makes the /t/ sound. Like as in top or tea. Oh, that's a bad example. That would be really confusing to the kids. I meant tea. They don't know that. Anyway, I would say two words that start with a t sound. And then I might have them go back to their desks and we'll practice drawing the letter ‘T’ and doing all those other things. All of that would happen way before any kind of productive failure.

My thoughts:

Burns rightly flags "tea" as a poor example here because it risks conflating letter names with sounds—a common pitfall in phonics. Words like "tea" start by using letter names as in a, b, c, and not letter sounds like /a/, /b/, /c/. The word "tea" is pronounced /ti:/, which is precisely the letter name for T ("tee"), so it doesn't serve as a clear exemplar for the isolated /t/ phoneme in a simple word. Broader examples abound where words begin with the pronunciation of letter names, potentially muddling the distinction for beginners: "bee" (/bi:/, starts with B's name "bee"), "ceiling" (starts with "see" like C's name), "deep" (starts with "dee" like D's name), "giraffe" (starts with "gee" like G's name), "item" ( starts with "eye" like I's name), "Jay" (starts with J's name "jay"). When teaching isolated sounds like /t/, stick to unambiguous exemplars (e.g., "top," "tap") to reinforce phonemic purity without introducing orthographic noise. This precision prevents the kind of early errors that Munger's "egg" analogy warns against, where the first flawed input seals off better alternatives.

 

Extract from the video:

And keep in mind it took me 10 seconds may be 20 seconds to do all that. And there would have been much better initial learning.

When you're doing initial learning, the best way is to model it. Provide lots of practice. Productive struggle is way down the road once they learn how to generalize it.

My thoughts:

Precisely—initial learning is paramount, as Thorndike articulated over a century ago in his foundational work on educational psychology.

The efficiency of explicit modeling (mere seconds for core introduction) allows immediate scaffolded practice, building toward automaticity far more effectively than drawn-out struggle. I'd advocate deferring any form of productive struggle until grade 7 and above, where learners have the metacognitive maturity and foundational skills to extract value from failure without derailing progress. Even then, it should be tightly structured, not a free-for-all, to ensure gaps are bridged promptly with targeted instruction.