Raw, Cooked, and Miscounted: Where Your Macros Drift

Grill a skinless chicken breast and, by USDA’s figures, its protein rises from 22.5 g to 30.5 g per 100 g ^{[1, 2]}. That is concentration, not added protein: water falls from 73.9 g to 66.1 g per 100 g, so roughly the same protein ends up in a lighter piece of meat.

Ground beef goes a step further. Pan-browned 80/20 crumbles list 272 kcal per 100 g, more than the 254 kcal of the raw meat ^{[3, 4]}, yet the meat lost fat as well as water ^[5]. Per 100 g, energy went up. For the meat on the plate, it went down.

None of these numbers is wrong. Each describes concentration in one food in one state; the trouble starts when a weight taken in one state is logged against an entry for another.

The actual question

The question is about measurement, not physiology: if you weigh food raw and log it as cooked, or the reverse, how far off is the log, and how much does the cooking method matter? We found no peer-reviewed study measuring how often people make this mismatch or how large the error is in real logs.

What exists is guidance from those who build food-composition tables. The FAO/INFOODS food-matching guidelines grade raw-food values applied to cooked foods without adjustment as a low-quality match that produces major under- or overestimates of intake ^[6]. In one example, pan-fried tomato matched to raw or boiled tomato comes out too low, because frying drives off water and adds oil.

How the database entries are built

The values here come from SR Legacy, the final edition of USDA’s National Nutrient Database for Standard Reference. Released in April 2018, it has been part of FoodData Central (FDC) since 2019 and is no longer updated ^{[1, 7]}. Entries report nutrients per 100 g of edible portion and mix laboratory analyses with calculations. The cooked 80/20 ground beef entries are labeled as analytical data derived by linear regression, from a study that modeled nutrients against fat content ^{[4, 5]}.

Raw and cooked versions of a food are separate records, not necessarily from the same samples or even the same decade. The raw and grilled skinless chicken breast entries each rest on six data points for water, protein and fat, both updated in 2013 ^{[1, 2]}; the roasted breast entry rests on 1979 data ^[8]; the rice entries date from 1989 and draw on different numbers of data points ^{[9, 10]}. Even for same-year pairs, FDC does not say whether the same pieces were analyzed before and after cooking.

Because analytical data for cooked foods are generally unavailable, most databases calculate cooked values from raw ones using cooking yields: hot cooked weight as a percentage of raw weight, reflecting moisture lost, water absorbed and fat gained or lost ^[5]. USDA’s retention formula, used for vitamins and minerals, states the key distinction: the amount retained is concentration multiplied by the food’s weight, before versus after cooking ^[11]. Concentration and total can move in opposite directions.

Energy is calculated too, from protein, fat and carbohydrate with conversion factors. FDC lists specific factors for some entries (4.27 kcal per gram of protein and 9.02 per gram of fat for raw chicken breast) and none for others, including grilled breast ^{[1, 2]}.

Concentration is not quantity

Muscle is roughly three-quarters water, and cooking mostly removes water: FSIS lists white-meat chicken with skin falling from about 69% to 61% water when cooked ^[12].

Read down the protein column and every cooked entry looks like a better protein source. Mostly it is the same food with less water, though the fried entry also lists the most fat and some carbohydrate, consistent with frying fat or a coating ^[14]. The two 2013 cooked entries differ by less than 2 g of protein per 100 g; the gap of about 3 g between grilled and fried also spans eras and product descriptions.

In ground beef, the higher figure per 100 g comes with less on the plate: USDA drained its browned crumbles in a colander, and its measured yields show both water and fat lost ^[5]. Drippings stirred into a sauce are still eaten.

Three illustrative calculations

These examples are arithmetic on the cited FDC values, as displayed on the FDC site. Each cooked weight is implied, not measured, by assuming one component, usually protein, survives cooking unchanged. That is our simplifying assumption, not a USDA method, and the answer shifts with the component chosen.

Why no single yield is right

In USDA’s table, medium-fat ground beef keeps 67% to 73% of its raw weight depending on method, losing about 21% to 28% of raw weight as moisture and 5% to 6% as fat. Low-fat beef keeps more (69% to 77%); high-fat beef keeps less (62% to 69%), losing about 12% to 13% as fat ^[5]. The Beef Checkoff Program funded the ground beef study.

Time and temperature matter too. In a laboratory experiment on 70 broiler breast muscles (10 per condition), cooking loss ran from 6.58% after one hour of sous-vide at 60 °C to 25.3% after three hours at 70 °C, and 26.5% in a convection oven at 180 °C ^[15]. Both factors had clear effects (P < 0.001); the absence of a significant interaction does not prove they simply add up. These laboratory figures show the size of the range, not a yield to use. FSIS adds that higher temperatures drive off more moisture, and that a beef roast cooked hot or long can lose about a third of its weight ^[12].

Cut matters as well, and the data run thin. USDA’s yields table gives roasted chicken with skin 72% for breast, 69% for thigh and 76% for drumstick, all from 1975, and has no row for skinless, boneless breast ^[5].

Water can also arrive before cooking. FSIS notes that retained-water declarations of 8% to 12% are not unusual for poultry, and that an added solution must appear in the product name and counts toward net weight ^[12]. FDC’s raw breast with added solution lists 20.3 g protein and 173 mg sodium per 100 g, against 22.5 g and 45 mg for the plain entry ^{[16, 1]}. US rules also let single-ingredient raw poultry that is not ground or chopped declare nutrition as consumed rather than as packaged ^[17].

What the numbers can and cannot show

All of this is arithmetic on reference data and inherits its limits. Gaps between entries are not clean causal effects: raw versus roasted chicken reflects cooking, but also samples analyzed decades apart. The sous-vide experiment supports a causal reading, but only within its controlled setup ^[15].

Most important, a logging error in one meal says nothing about results over a training block, and we found no research on whether these mismatches change body-composition outcomes. A reasonable inference, not a finding, is that a mismatch repeated daily, such as always logging cooked rice against the raw entry, is systematic and will not average out. SR Legacy is also frozen; newer FDC data types may list different values, so the arithmetic carries over but the exact numbers may not ^{[1, 7]}.

Making the log match the plate

The arithmetic is solid; its effect on results is not. A few habits are a proportionate response:

• Weigh and log in one state. Raw weight goes with a raw entry, cooked weight with a cooked entry for the same method.
• Use raw totals for batch cooking. Log the raw ingredients, weigh the cooked batch and assign portions by cooked weight; for drained fatty meat, raw totals overstate fat.
• Weigh grains and note the state. A cup of dry rice carries more than three times the energy of a cup of cooked rice ^{[9, 10]}.
• Log what you add. Oil, butter, marinades, breading and sauces made from drippings.
• Read the package. Look for added solution, retained water and whether the panel is as packaged or as consumed ^{[12, 17]}. If you are on a sodium-restricted diet for a medical reason, ask your clinician about added-solution products.

A database entry answers a narrower question than a food log asks: how much of each nutrient is in 100 g of this food, in this state, prepared this way. A log is only as accurate as the match between the weight and the entry.