Your Favourite Show Probably Didn’t Decline

What 192,720 episode ratings say about television’s most durable complaint

Every number in this piece is backed by the analysis logged in decisions.md. The formal write-up, with methods and full results, is in technical_report.md.

1. The show that stayed too long

Think of a show you loved that stayed on too long.

Got one? Now the uncomfortable question — and be honest, because your memory has a stake in the answer. Was that show actually worse by the end? Did the writing genuinely rot, the characters flatten, the spark drain out of it? Or did the ending simply let you down, and does a fresh disappointment shout louder than the dozens of quiet, good episodes that earned your love in the first place? We are, all of us, far better at remembering how a story left us than how it treated us along the way.

Whatever you answered, you did not answer alone. We all carry the same conviction, worn so smooth by repetition that it passes for common sense: television declines. Shows arrive brimming with promise, spend their best ideas too early, and coast downhill to a finish nobody asked for. Game of Thrones, Dexter, The Walking Dead, pick your own heartbreak. It reads less like an opinion than a law of the medium.

So we put it on trial. We pulled the episode ratings from IMDb, 192,720 episodes across 3,234 shows, and traced where each show’s ratings actually travel, from premiere to finale.

The law is wrong.

2. Most shows just don’t decline

Draw a straight line through a show’s episode ratings, first to last, and you have its whole life in a single number: how much it gained or lost between the pilot and the finale. Do that for all 3,234 shows and the belief falls apart on the very first count:

• A clear decline, meaning the show gives up at least half a rating point across its whole run, turns up in just 16.9% of shows. About one in six.
• The largest group by far, 57.4%, is essentially flat: no real rise, no real fall.
• 25.7% clearly rise, judged by that same half-point bar. Climbers outnumber decliners three to two.
• The median show doesn’t sag at all. It drifts slightly up. Ignore the size of the move and simply count directions, and it still tips that way: for every three shows that drift down, four drift up.

Figure 1. Every show’s trend, first episode to last. The distribution is a hill sitting just to the right of zero. The declining tail is real, but it is a tail.

If shows really coasted downhill, we’d expect a room full of falling lines. Instead most of them hold their ground, and more climb than fall.

And they hold on more stubbornly than you’d guess. Split any show down the middle and compare its first half to its second: the two track each other almost perfectly, a +0.79 correlation. Even the shows that open strongest, the top quarter averaging an 8.3, surrender a mere 0.07 of a point by their back half. Whatever drama plays out inside a single season, a show’s overall quality barely budges from beginning to end.

Figure 2. Each show’s first half plotted against its separately measured second half, with no shared episodes, so nothing is being compared with itself. The cloud hugs the line of no change, and the panel beside it shows how little even the extremes move.

Which leaves a puzzle. If the numbers are this clear, why is everyone — including you, thirty seconds ago — so sure that shows fall apart?

That gap is the real story. Here’s where it comes from.

3. The finale is a victory lap, not a car crash

If a show is going to break your heart, you know exactly when it happens: the end. That is where Game of Thrones detonated, where Dexter became a punchline, where a decade of goodwill goes to die. So the finale should be the low point, the moment the long decline finally lands.

It’s the high point.

Go season by season through all 3,234 shows and the season finale is, more often than not, the best episode in its season. It beats the rest of its own season 72.6% of the time, by a quarter of a point on average. Shows save their strongest hour for last. And, tellingly, they don’t do the reverse: the season premiere is thoroughly unremarkable, no better than some random mid-season episode. Television builds toward endings; it doesn’t coast away from beginnings.

A show’s very last episode plays by the same rule. It lands above its final season’s average 71.6% of the time. Only about one in ten finales falls hard enough to count as a real collapse.

Figure 3. Season finales beat the rest of their own season nearly three times out of four; premieres are unremarkable by comparison. The very last episode a show ever airs behaves like every other finale: it goes out high.

That one in ten is simply the one you remember. Dexter, How I Met Your Mother, Power, You, Veronica Mars — the infamous flops are real, and the data registers every one of them. The last of How I Met Your Mother ran around a 7.6, and then it signed off at a 5.5. But for every one of those, there’s a handful that leave the way The Office, The Big Bang Theory, Brooklyn Nine-Nine, and Star Trek: The Next Generation did, each closing on an episode rated a full two points above the season around it, the audience on its feet.

The finale you dread is the exception. The finale as a curtain call is the rule.

4. The curse that only strikes the famous

Fine. A finale is one episode. Maybe the rot is bigger than that; maybe it’s the whole final season that goes soft, the writers spent, the budget stretched, the story limping to its contractual end. That is the real final-season curse, and Game of Thrones is its patron saint: a last season averaging a 6.4 against a 9.0 for everything that came before.

To test it we took every show that has actually finished and set its final season against the rest of its run. The result comes out almost exactly even. 50.5% have a weaker final season, 49.2% a stronger one, and the middle of the pack moves by essentially nothing. There is no curse. On average, a final season is just another season.

But a genuine collapse does strike a small minority, and this is where the belief gets its oxygen. About one in eight shows (12.6%) sheds half a point or more at the end. Look at who they are: Game of Thrones, down 2.5 points; Master of None, down 2.5; The Promised Neverland, down 2.3. These are the collapses that launch a thousand angry threads. The curse is real. It is just rare — and it keeps landing on the shows people actually cared about.

Figure 4. Every ended show’s final season, set against the rest of its run. The bulk of the distribution sits on zero; the curse is that red left tail, and it stays a minority wherever the line is drawn.

Which might be the entire trick. And this part we can actually check.

Sort every show by how many people voted on it, the best proxy we have for how many watched. The belief gets truer as the audience gets bigger. Among the most obscure shows in the data, 15% clearly decline. Among the sixty household names, the ones carrying more than half a million votes, it is 43%. Nearly half. The final-season curse tracks the same way: 12% for the quiet shows, 24% for the famous ones. The climb is smooth at every step, and we tried three ways to break it: that it was simply noisier data on obscure shows, that it was the length effect wearing a disguise, that it was age and length together. It survived all three.

So the belief isn’t stupid. It is an accurate description of a different dataset. No group chat ever erupted over a quiet drama that slipped a point in season four; the shows you have watched all the way to the end are the famous ones, and the famous ones really do decline. You never formed your opinion from a random sample of television. Nobody has.

One honesty note before we move on. We can’t point the arrow here. A show that ends badly is precisely the show people pile in to rate, so fame and disappointment feed on each other, and this data can’t say which came first. What it can say is descriptive, and it is enough: “about one in six” was never a statement about your shows.

5. Decline is real — just not where you’re looking

We have spent four sections tearing the belief down. Time to be fair: decline is real. It simply isn’t lurking where you’d look for it, and it has almost nothing to do with a show “running out of ideas.” It comes down to two unglamorous facts about a show — and, if you insist, a genre.

The first is length. The longer a show runs, the more it wears down. Series that wrap up in two or three seasons mostly hold steady or climb; among shows that grind on for six seasons or more, a clear majority, 61%, decline. Every extra season tilts the odds a little further downhill. Which quietly reframes the whole story: many of the shows we remember as “declining” are really just the shows that would not end, the ones that outran their story and kept airing anyway.

The second is age. Older shows sink; newer ones rise. Series that premiered before 2000 mostly trend downward over their run, while nearly everything since trends up. The split holds even after you account for length, so it isn’t simply that old shows ran longer. Why is harder to answer, and here we are guessing rather than measuring. The modern hit is short, serialized, and plotted end to end, engineered to build toward a payoff. The classic network show was episodic and open-ended, built to run forever. Forever tends to plateau, then sag.

And genre? Mostly a red herring. Hold age and length constant and the giants of the schedule, drama and comedy and crime and action, come out flat, none of them cursed or blessed by their label. Testing sixteen genres at once means roughly one will look meaningful by pure luck, so we raised the bar to account for that, and only two categories clear it: animation quietly climbs, those long-running cartoons and anime that sharpen with age, while adventure drifts the other way. Documentary, romance and thriller probably belong on that list too, on a slightly more forgiving test. Biography looked like a decliner until we applied the correction, and then it didn’t.

Figure 5. One regression, every factor at once, so each effect is what remains after the others are held fixed. Length and age each survive that test. Almost no genre does: grey marks the estimates that fail to clear the bar once you account for testing sixteen genres at once, and that is fourteen of the sixteen.

“Do shows decline?” finally has an honest answer. Some do, but less because they got worse than because of how old they are, and how long they refused to stop.

6. Nobody peaks where you think

Two smaller myths travel with the decline story, and the data retires both.

Start with the louder one: that a show enjoys a “golden age” early and it is all downhill from there. Here the data half-agrees, and it’s worth being precise about how. The single most common trajectory, about 45% of shows, really is an arc: a climb to a peak, then a drift back down. It is the closest thing in the numbers to “jumping the shark.”

Figure 6. The rise-then-fall arc is the single most common shape a run takes. Look at the vertical scale on the right, though: the entire arc plays out inside a few tenths of a rating point.

But two facts drain the melodrama out of it. First, the peak isn’t early. On average it lands near the middle of a show’s run, and it slides later the longer the show goes: a three-season show peaks around its second year, a long-runner somewhere in its middle years. (That famous “second-season surge”? Mostly an illusion created by how many shows are short.) Second — and this is the part that matters — the arc is gentle. Recall the very first thing we found: the average show ends within a whisker of where it began. The hump is real. It’s a hill, not a cliff.

Chasing that peak retires the second myth almost in passing: the first season is always the best. It usually isn’t. Across our shows, the single strongest season is the last one more often (37%) than the first (27%). A show is likelier to save its best for the end than to open on it.

Figure 7. Where a show’s best season falls. The peak drifts later the longer a show runs (left), yet it is not concentrated in the middle (centre), and the first season takes the crown less often than the last (right).

So yes, shows tend to rise, peak, and ease back down. They just peak later, and slide far less, than the legend swears they do.

7. How we kept from fooling ourselves

A conclusion this far from common sense deserves suspicion — including our own. We spent as much effort trying to break these findings as making them. Here are the traps we had to climb back out of. Each one is the difference between a real result and a confident mistake.

What counts as a show. Three IMDb tables, joined so that every row is one rated episode of one series. Then a bar to clear, because a trajectory needs enough show to have one: at least 13 rated episodes, at least two seasons, and an average of at least 50 votes per episode, so we’re reading signal rather than noise. Game shows, reality TV, talk shows and news are excluded outright: “decline” is a claim about a story, and those formats aren’t telling one. What survives is 3,234 shows and 192,720 episodes, everything from two-season miniseries to the 806 episodes of The Simpsons.

That population is the whole essay, and behind every percentage in it sits a picture like this one: a few dozen episode ratings and a line drawn through them.

Figure 8. One show from each of the three groups in Figure 1. Every dot is an episode, sized by how many people voted on it; the line through them is the single number the show gets scored on. Friends is what “essentially flat” looks like across ten seasons.

Whose votes count. Not every episode rating is equally trustworthy: a finale rated by 300,000 people is a firmer number than a mid-season hour rated by 80. But raw vote counts are their own trap, because a show’s most-voted episodes are often its most notorious ones. Four of Game of Thrones’ six most-voted episodes come from its final season, every one of them rated far below the show’s own average: an audience that turned out mainly to register its disappointment. Weight by raw votes and that late cluster hijacks the entire trend. We settled on the square root of the vote count, which trusts popular episodes more without letting one viral meltdown speak for a decade.

Figure 9. The same 73 episodes, three ways of counting them. Weighting by raw votes gives the steepest line of the three: the final season’s backlash drags the whole decade’s trend down with it.

The sentence we got wrong. A smaller confession, and the reason we re-run everything. That Game of Thrones observation first went into our notes as “its two most-voted episodes are its last two”, a tidy line that made the point beautifully. It was also false. The show’s single most-voted episode is the big dot near the top of Figure 9: a mid-run 9.9, one of the most beloved hours it ever aired. We caught it only by finally running a script we had written and never actually executed. The conclusion held; the sentence propping it up did not. Headline findings get scrutinised. It’s the confident little sentences around them where mistakes survive.

The trap we nearly fell into. Early on, a number told us high-rated shows decline hard: a strong −0.37 link between where a show starts and how it trends. It looked like a headline. It was an artifact. We had correlated each show’s trajectory with its own starting point, and when one estimate feeds both sides of a comparison, it manufactures a relationship out of thin air. The honest version compares a show’s first half against its separately measured second half (Figure 2), and there the effect nearly evaporated: the highest-starting shows give up all of 0.07 of a point. A finding we would have happily published was mostly a mirage. It is why we now distrust any single number until it survives a cleaner test.

Untangling the culprits. “Newer shows rise,” “shorter shows rise,” “older shows decline” sound like three separate discoveries, but newer shows are shorter, so they might be one discovery in three costumes. We pulled them apart with a model that holds the others fixed, and both age and length survived on their own. Two real forces, not a single confound in disguise.

How firm is “16.9%”, really? Each show’s trend is measured, not known, so a show sitting near the line could fall on either side of it by luck. At the level of one show that bites: half our “clear decliners” can’t be shown to have lost a full half point rather than simply lost ground. Across 3,234 shows it mostly cancels, and what doesn’t cancel works against us rather than for us, because the flat middle is the only group with two edges to leak across. Correcting for it moves the headline down about a point, not up.

Where the data thins out. One pattern — shows that dip and then “find themselves” again — looked solid until we noticed that a show’s late episodes are precisely its least-voted ones (audiences thin out; only the diehards keep rating). For about a fifth of those shows, the late “recovery” leaned on a handful of low-vote episodes and dissolved the moment we demanded firmer numbers. Anything resting on a show’s final stretch is now treated as the shakiest ground in the dataset, and labelled as such.

None of these checks moved the headline. They changed how much we trust it, which is exactly the point.

8. What the numbers can’t see

Every finding here rests on one blunt proxy: a show’s IMDb episode ratings. That buys us 192,720 data points, and a few blind spots worth naming plainly.

A rating never says why. Our method sees a show’s score fall; it has no idea what pushed it. And the push often comes from off-screen, not from the writing:

• House of Cards has the steepest final-season crash in the whole dataset, down 4.2 points, far past the 2.6 that made Game of Thrones infamous. Its lead actor was written out amid scandal and the last season was rebuilt without him.
• Top Gear didn’t run out of ideas either. Its star, Jeremy Clarkson, punched a producer in 2015 and was fired; his two co-hosts left with him, and what followed shows up in our numbers as a two-point drop, one of the largest we measured.
• Two and a Half Men lost its lead the same way, when Charlie Sheen was fired mid-feud, and no amount of Ashton Kutcher held the room.
• Scrubs “declined” in its ninth season only because it had quietly become a different show: new cast, new setting, a soft reboot wearing the old name.
• Even Game of Thrones, patron saint of the decline story, has an off-screen culprit. It sprinted past the novels it was adapting and wrote its final seasons without a map.

To our numbers, these are all just “decline.” A fired actor, a punched producer, and a novelist who hadn’t finished the books look identical to a show that simply lost its spark. We can measure the fall; we can’t read its cause.

The crowd isn’t the audience. IMDb scores come from whoever chose to log in and vote: a self-selected, disproportionately online crowd, not a representative sample of viewers. And that crowd organises. Some of the lowest-scored episodes in our data were rated down in anger, review-bombed by fans lodging a protest rather than an opinion. Weighting by vote count blunts this; it doesn’t erase it.

A few formats slipped the net. Our filters aim at scripted, narrative shows, but a handful of sketch and variety programs wear the “comedy” label and sneak through. Saturday Night Live, a thousand-plus episodes of it, is not the season-arc story we set out to measure. A rounding error rather than a distortion, but it’s there.

None of this sinks the headline: most shows still don’t decline, and the finale is still usually a peak. It just draws the honest border around what an analysis like this can claim. We can map where television ratings move, and how much. The why belongs to a messier world the numbers can’t see.

9. The show you were thinking of

Back to your show, then. The one you picked in the first paragraph.

It might genuinely have fallen apart. One in six do, and if you picked a household name, closer to two in five — section 4 was honest about that, and it means this is a real possibility rather than a long shot. But the odds are still against it, and there’s a fair chance your show is Lost.

If any series is shorthand for betrayal, it’s that one: six seasons of questions, an ending people have been arguing about since 2010, the standing punchline for a show that lost the plot. Here is what its ratings actually did.

Season one averaged 8.54. Season six averaged 8.32. That is the entire collapse: 0.22 of a point. Two consecutive Lost episodes differed by more than that on an ordinary week, 69% of the time.

And that worst-ever season still beats the best season three out of four shows in our data ever managed. Across all 121 episodes the trend line is flatter than 98% of everything we measured; of the 499 shows that ran past a hundred episodes, exactly twelve hold a straighter one. The pilot scored a 9.1. The finale that supposedly ruined everything is rated 9.2 today, the eighth-best hour the show ever aired.

Lost never declined. We only remember it that way.

Which is the whole finding, in one show. The disappointment is real — but it is a fact about endings, not about series.

We are, all of us, far better at remembering how a story left us than how it treated us along the way.