Nutrition quality of food purchases varies by household income: the SHoPPER study

This was an observational study comparing the nutritional quality of household food purchases across different income levels. The researchers did not test an intervention — they simply collected and analysed food purchase receipts from 202 urban households over 14 days. The main comparison was between households with different income-to-poverty ratios (how close a household's income is to the federal poverty line). The primary outcome was the Healthy Eating Index 2010 (HEI-2010) total score, which measures how well a set of foods aligns with U.S. dietary guidelines (0–100 scale, higher = healthier). Secondary outcomes included HEI sub-scores for specific food groups (vegetables, fruits, dairy, whole grains, etc.) and the proportion of grocery dollars spent on specific categories like sugar-sweetened beverages, frozen desserts, and snacks.

**Sample size:** 202 households (each household contributed one set of purchase data)

**Population:** Urban households in the Chicago metropolitan area, recruited through a combination of community advertising, flyers, and word-of-mouth

**Inclusion criteria:** Households had to do at least half of their food shopping at a grocery store (not restaurants or convenience stores), speak English, and have at least one adult aged 18–65 who was the primary food shopper

**Exclusion criteria:** Households that did not save receipts or that primarily shopped at farmers' markets or specialty stores

**Demographics:** Mean household size was 2.5 people; 68% of primary shoppers were female; 45% were non-Hispanic white, 38% non-Hispanic Black, 12% Hispanic; median annual household income was $40,000 (range: <$10,000 to >$100,000); 28% of households had income below the federal poverty line

**Setting:** Participants' homes and local grocery stores in the Chicago area

**Food purchase data:** Households collected all food and beverage purchase receipts for 14 consecutive days. Receipts included grocery store purchases, convenience store purchases, and any other food bought for home consumption (but not restaurant meals or takeout eaten away from home). Participants were given a logbook to record any purchases where the receipt was lost.

**Nutritional analysis:** All food items on receipts were entered into the Nutrition Data System for Research (NDS-R) software, which calculates nutrient content and food group servings based on standardised portion sizes. This is the same software used in many clinical nutrition trials and is considered a gold standard for dietary assessment.

**Diet quality scoring:** The NDS-R output was used to calculate the Healthy Eating Index 2010 (HEI-2010), which has 12 components: total fruit, whole fruit, total vegetables, greens and beans, whole grains, dairy, total protein foods, seafood and plant proteins, fatty acids, refined grains, sodium, and empty calories (from solid fats, added sugars, and alcohol). Each component is scored 0–5 or 0–10, with a maximum total of 100.

**Income measurement:** Household income was reported as a categorical variable (e.g., <$10,000, $10,000–$19,999, etc.) and converted to an income-to-poverty ratio using federal poverty thresholds adjusted for household size. Households were then divided into three groups: low income (ratio <1.85, meaning income less than 185% of the poverty line), middle income (ratio 1.85–3.00), and high income (ratio >3.00).

**Covariates:** Education level of the primary shopper, marital status, and race/ethnicity were self-reported and included as statistical controls.

**Study design:** This is a cross-sectional observational study. There was no randomisation, no blinding, no intervention, and no control group. The researchers simply measured food purchases and income at one point in time and looked for associations.

**Why this design was chosen:** The study was designed to describe real-world purchasing patterns in a diverse urban sample. A cross-sectional design is appropriate for generating hypotheses about income-diet relationships and for identifying potential targets for future interventions. It is not designed to prove that income *causes* differences in diet quality — only that they are associated.

**Data collection procedure:** Each household was given a receipt collection kit (envelopes, logbook, instructions) and asked to save every food and beverage receipt for 14 days. Research staff called participants twice during the collection period to remind them. After 14 days, staff visited the home to collect receipts and logbooks. Receipts were then entered into NDS-R by trained research assistants. Quality control checks were performed on a random 10% of receipts.

**Statistical approach:** The primary analysis used linear regression to compare HEI total scores across income groups, adjusting for education, marital status, and race. Secondary analyses used the same approach for HEI sub-scores and for proportion of grocery dollars spent on specific food categories. P-values were reported as adjusted (from the regression model) and unadjusted. No correction for multiple comparisons was applied to the secondary outcomes, which increases the risk of false positives.

**What this design can prove:**

That there is a statistical association between household income and the nutritional quality of food purchases

The magnitude of the difference between income groups

Which specific food categories differ most between income groups

**What this design cannot prove:**

That income *causes* differences in diet quality (reverse causation is possible — e.g., people who buy healthier food may earn more money due to better health)

That the observed differences are due to food access, knowledge, preferences, or any other mechanism

That the results apply to non-urban populations, households that eat many meals away from home, or households that primarily shop at non-grocery stores

**Major methodological weaknesses:**

**Self-reported receipt collection:** Households may have forgotten to save receipts, leading to incomplete data. The logbook was meant to capture missing receipts, but this still relies on memory.

**14 days is a short window:** Food purchases can vary seasonally, weekly, or due to special occasions. A single 14-day period may not represent typical purchasing patterns.

**No restaurant or takeout data:** The study only captured food bought for home consumption. Lower-income households may eat more fast food or takeout, which would not be captured here. This could underestimate the true diet quality gap.

**No individual-level dietary intake:** The study measured what households *bought*, not what individuals *ate*. Food could be wasted, shared with non-household members, or prepared in ways that change its nutritional value.

**No blinding:** Researchers knew which households were in which income group when entering and analysing data. While NDS-R entry is somewhat objective, there is still potential for bias.

**Multiple comparisons:** The authors tested many sub-scores and food categories without adjusting for multiple comparisons. Some of the "significant" findings may be due to chance.

**Primary outcome — HEI total score:** Higher-income households had a mean HEI total score of 68.2 (SD = 13.3) compared to 51.6 (SD = 13.9) for lower-income households. This difference of 16.6 points was statistically significant after adjusting for education, marital status, and race (adjusted p = 0.05). Middle-income households scored 60.1 (SD = 14.2), which was not significantly different from either extreme group after adjustment.

**Vegetable sub-score:** Higher-income households scored 3.6 (SD = 1.4) out of a possible 5, compared to 2.3 (SD = 1.6) for lower-income households (adjusted p < 0.01). This was the largest proportional difference among sub-scores.

**Dairy sub-score:** Higher-income households scored 5.6 (SD = 3.0) out of 10, compared to 5.0 (SD = 3.3) for lower-income households (adjusted p = 0.05). The difference was small but statistically significant.

**Frozen desserts spending:** Lower-income households spent 3% (SD = 7%) of their grocery dollars on frozen desserts, compared to 1% (SD = 2%) for higher-income households (p = 0.02). This was the only food category that showed a statistically significant difference in spending proportion.

**Non-significant findings:** No significant differences were found for total fruit, whole fruit, whole grains, total protein, seafood/plant protein, fatty acids, refined grains, sodium, empty calories, or spending on sugar-sweetened beverages, snacks, or vegetables (as a proportion of spending). The lack of significance for sugar-sweetened beverages is notable given that this is often assumed to be a major driver of income-based diet disparities.

**Covariate effects:** Education, marital status, and race were all independently associated with HEI scores, but the income effect remained significant after controlling for them. This suggests income has an independent association with diet quality beyond these demographic factors.

The 16.6-point gap in HEI scores between the lowest and highest income groups is substantial. To put this in context: the average American adult has an HEI score of about 59 out of 100. A 16.6-point difference is roughly equivalent to the difference between someone who eats fast food twice a week and someone who eats fast food once a month, or between someone who eats vegetables at one meal per day versus three meals per day. The vegetable sub-score difference (1.3 points out of 5) means lower-income households are buying about 26% fewer vegetable servings per dollar spent. The frozen desserts finding means lower-income households are spending three times as much of their grocery budget on ice cream and similar products — but since the absolute percentage is small (3% vs. 1%), this translates to roughly $2–$4 more per week for a typical household.

**Author-acknowledged limitations:**

The sample was limited to urban households in one metropolitan area (Chicago), so results may not generalise to rural or suburban populations

The 14-day collection period may not capture seasonal or occasional purchases (e.g., holiday foods, bulk buying)

Receipt data may be incomplete if participants forgot to save receipts

The study did not measure actual dietary intake, only purchases

The income-to-poverty ratio is a crude measure that does not account for cost of living differences or household debt

**Additional critical limitations:**

**No adjustment for multiple comparisons:** The authors tested dozens of sub-scores and food categories without correcting for the increased risk of false positives. Only the vegetable sub-score and frozen desserts spending survived even a lenient correction (Bonferroni would require p < 0.004 for 12 sub-scores).

**The primary outcome barely reached significance:** The adjusted p-value for the main HEI total score comparison was exactly 0.05 — the conventional threshold. This is borderline and could easily become non-significant with a slightly different model specification or a few more participants.

**No data on food access:** The study did not measure proximity to grocery stores, availability of healthy options, or transportation barriers. These factors could confound the income-diet relationship.

**Self-selection bias:** Households that agreed to participate in a study about food shopping may be more health-conscious or organised than the general population, potentially biasing the results.

**No data on food waste:** Higher-income households might buy more vegetables but waste more of them, meaning the actual intake difference could be smaller than the purchase difference.

**Industry funding:** The study was funded by the National Institutes of Health (NIH), so no direct industry bias. However, the NDS-R software is developed by the University of Minnesota, which has received funding from food industry sources in other studies.

For someone running their own n=1 experiment on food purchasing and diet quality:

**What to test:**

Test whether increasing your vegetable purchases (by volume or variety) improves your overall diet quality as measured by a simplified HEI score

Alternatively, test whether reducing frozen dessert purchases (or replacing them with fruit-based desserts) changes your daily calorie intake or vegetable consumption

A more ambitious test: compare two weeks of "budget-conscious" shopping (spending as little as possible) versus two weeks of "quality-focused" shopping (spending without budget constraints) on the same HEI metrics

**Minimum meaningful duration:**

14 days minimum (matching the study's collection period) to capture typical weekly variation

28 days (two full months) would be better to account for biweekly pay cycles, pantry depletion, and occasional purchases

For a true n=1 experiment, aim for 4–6 weeks per condition to see stable patterns

**What to measure:**

**Primary metric:** Simplified HEI score for your purchases. You can calculate this manually using the HEI-2010 scoring guidelines (available from the USDA) or use a food tracking app that provides HEI estimates. Focus on the 12 components: total fruit, whole fruit, total vegetables, greens and beans, whole grains, dairy, total protein, seafood/plant protein, fatty acids, refined grains, sodium, and empty calories.

**Secondary metrics:** Proportion of grocery dollars spent on vegetables, fruits, sugar-sweetened beverages, and frozen desserts. Track total weekly grocery spending. Also track your actual dietary intake (using a food diary or app) to see if purchase changes translate to intake changes.

**Confounders to measure:** Total grocery budget, number of people eating from the household food supply, number of meals eaten away from home, time spent cooking, and any special events (holidays, parties, guests).

**Key confounds to control for:**

**Total spending:** If you change your budget, you change what you can buy. Keep total grocery spending constant across conditions, or at least track it.

**Household size:** If you live alone, your purchasing patterns will differ from a family. Note how many people you're shopping for.

**Seasonality:** Vegetable prices and availability vary by season. Run your experiment in the same season or across multiple seasons.

**Cooking time:** More nutritious food often requires more preparation time. Track time spent cooking and meal planning.

**Food waste:** Weigh or estimate food waste (especially vegetables) to see if you're buying more than you eat.

**Stress and mood:** Emotional state affects food choices. Keep a brief daily mood log.

**Shopping frequency:** Shopping daily vs. weekly vs. monthly changes what you buy. Keep frequency consistent.

**What a positive result would look like:**

A 10-point or greater increase in your simplified HEI score (from, say, 55 to 65) after changing your purchasing patterns

A 20% or greater increase in the proportion of grocery dollars spent on vegetables (e.g., from 5% to 6% of total spending)

A 50% reduction in frozen dessert spending (e.g., from $5/week to $2.50/week)

A measurable improvement in your actual dietary intake (e.g., 1–2 more servings of vegetables per day) that persists for at least two weeks

A positive result would be most convincing if you see it in both your purchase data and your intake data, and if it replicates across two separate trial periods

**Watch out for:** The biggest confound in this study was that the researchers measured purchases, not intake. In your own experiment, you might buy more vegetables but not eat them (food waste), or you might buy fewer frozen desserts but compensate by eating more restaurant desserts. Always measure actual intake alongside purchases. Also note that the study found the biggest income-related difference in vegetables, not in sugar-sweetened beverages or snacks — so if you're trying to improve your diet quality, focusing on vegetable purchases may be more impactful than cutting out treats.