Household Cost Methodology

A complete derivation of the monthly household cost of incorporation — every input, its source, how much confidence it carries, and what would have to be true for the figure to be wrong. Written for a reader who wants to check the arithmetic rather than take it on trust. Last revised 9 October 2026.

The Figures

These are best estimates. A 200,000-draw simulation over the full range of every uncertain input puts the true cost to a median home at a median of $62.47, and the published $62.66 at the 52nd percentile of that distribution — as likely to land high as low.

The range is $55 to $71 a month. Everything that produces that width is set out below, input by input. Nearly four-fifths of that width is one input, the Census margin on Niwot’s own household income. The household-size split introduced on 16 September contributes 1% of it, despite resting on a cell with a 45% margin of its own — because it redistributes a total it does not change.

That interval holds one thing fixed, and this paper now says so. It assumes the method used to spread a fixed town-wide spending total across dwellings is the right one. There are two defensible methods and we use both — household size on this site, home value in the neighbourhood letters. Letting the choice between them vary as well widens the interval to $53 to $70 and moves the published $62.66 from the 52nd percentile to the 67th. See The choice of method is itself an uncertainty below. We publish the higher of the two figures our own models produce; on the full uncertainty there is roughly a two-in-three chance the published number is at or above the true one.

How to Read the Confidence Statements

There is no honest way to put a single ± figure on this number, and any document that offers one should be treated with suspicion. The uncertainty here is not sampling error around a measurement. It is a chain of five steps, some of which rest on published data and some on judgment, and those two kinds of doubt do not combine into one percentage.

So this document expresses confidence three ways, and a reader should use all three:

1. An evidence tier for every input — 1 through 4, describing what kind of support the number has, not how precise it is. A tier-1 input can still be the wrong input to use; a tier-4 input can happen to be right. The tier says how much argument is available when someone challenges it.
2. A modelled range for the output — from drawing every uncertain input across a stated range. This is not a statistical confidence interval, and calling it one would be a misrepresentation. It answers “how wrong could this be, given how wrong each input could plausibly be,” where the plausible ranges are our judgment except where noted.
3. A break-even table — for each input, how far it would have to move, on its own, to change the answer. This is the most useful of the three in an argument, because it converts every objection into a specific claim someone has to defend with a number.

The Chain

The derivation has five junctions. Two questions run through this material and confusing them produces most objections we receive: burden (what one household pays) and incidence (where the town's revenue comes from). They do not divide into one another. This document is about burden; incidence is treated at the end.

One structural point matters and is easy to get wrong. The business-intensity factor that scales the town's revenue does not enter the household cost at all. Household cost depends on the household share of the base, and on nothing about how many businesses Niwot has. An earlier version of this model multiplied the two together — effectively asserting that Niwot households buy less telephone service because Niwot has fewer offices — and that error understated household cost by about $4 a month until it was found on 12 September 2026.

Junction 1 — The Anchor

Colorado sources a municipal sales tax to the delivery address, so a town taxes everything delivered into it, not only what is bought at a shop inside it. No Colorado town publishes what its households buy that way. Louisville publishes the closest thing: its “Outside City” taxable base, broken out by industry, in the monthly Finance Committee packet.

The eleven-month problem, and why it does not matter

Grossing eleven observed months up by 12/11 assumes the missing month is an average one. It is not. Colorado publishes monthly taxable sales by city, so this can be measured rather than argued. Across ten complete years, Louisville's seasonal index runs:

December runs 16% above an average month. But which month is missing depends on a convention we have not confirmed — a city report labelled “Jan–Nov” may run on sales months or on collection months, and Colorado remits one month in arrears. Both cases were computed. If the months are sales months, ×12/11 understates the base by 1.45%; if they are collection months, it overstates by 0.35%. The error is bounded between those, which is under fifty cents a month on the household figure. The convention question therefore never needs answering, which is a better outcome than answering it.

Junction 2 — The Transfer

Louisville is a larger, differently-composed town. Carrying its figures to Niwot requires an income adjustment, and this is where most of the remaining uncertainty lives.

We work entirely in household terms on both sides. This matters: an earlier version scaled Louisville's per-capita sales by a per-capita income ratio and then multiplied by Niwot's persons-per-household — mixing a Census quantity measured on the Niwot census place with one built from the committee demographer's boundary population divided by a parcel count. Those come from different universes, and the two routes disagreed by 4 to 11%. The household route never uses per-capita figures at all.

Three income measures that disagree, and why we chose this one

A reader checking our work against Census tables will find three different answers to “how much richer is Niwot than Louisville,” and the choice is consequential:

On median household income the two towns are identical. The gap between the measures is entirely the shape of the income distribution: Niwot's is more top-heavy (mean 1.34 times median) than Louisville's (1.24).

We use mean household income, because the quantity being scaled is an aggregate taxable base, and an aggregate scales with the mean, not the median. Using per capita income instead would raise the answer by roughly $2 a month; using median household income would lower it by roughly $1.60. That spread — about $58 to $62 — is the honest width of this single choice.

Two things the income figures do not tell us

• Footprint. The Census measures the Niwot census designated place, which holds 3,447 people. The proposed boundary holds 4,305, per the committee's demographer, which supersedes the Census figure here. So 858 residents, a fifth of the town, are inside the boundary but outside the area the income figure describes. If their income differs from the rest by 25%, the answer moves about $1.10; if by 40%, about $1.80. A real bias, but a bounded and small one.
• Margin. Niwot is a small place and the Census sample there is thin. Mean household income is built from two estimates, each with its own margin: aggregate household income is ±$61,172,000 on $296,034,900 (20.7%), and the household count is ±209 on 1,505 (13.9%). Together they imply roughly a 15% coefficient of variation on the income ratio — and that single quantity is the largest contributor to the uncertainty in the published figure by a wide margin. No amount of care elsewhere in the model compensates for it. Per capita income, which we do not use, is measured more precisely (±$14,859, or 17.5% of a smaller number); it was rejected for the structural reason above, not the statistical one.

The elasticity understates, and we left it that way

Updated 28 September 2026: this exponent now has a sourced neighbour. See the note at the end of this section.

The 0.65 exponent has no citation. It can, however, be tested, though only weakly. Two towns are observed — Louisville and Lyons — with both their incomes and their remote taxable bases known. The elasticity that makes those two reconcile exactly is 1.42. This is a line through two points, and one of them sits on a different reporting basis (see Comparability Cautions below), so it bounds a direction rather than estimating a value. At that value the published figure would rise by about $1.20 a month. We publish the lower number. Nothing in the observed data suggests 0.65 is generous; it is the cautious end.

Since 23 September there is a fitted value for this quantity, and 0.65 sits above it. The elasticity of taxable spending to income, measured from the 2023 Consumer Expenditure Survey across five income quintiles and restricted to the categories Colorado taxes, is 0.549 at R² = 0.994 — a far better-evidenced number than either the uncited 0.65 or the two-point 1.42. Niwot’s income ratio to Louisville is 1.0792, so the exponent multiplies the transferred spending by 1.0508 at 0.65 and 1.0427 at 0.549: substituting the fitted value would lower the published figure by about $0.25 a month. The published exponent therefore states a slightly higher cost than the best-sourced one, which is the direction to err in, and it has been left where it is. The sentence above — that 0.65 has no citation — was true when written and is no longer the whole picture; it is left standing with this note beneath it rather than rewritten.

Junction 3 — The Household Share

The delivered-goods base is not all household shopping. It includes business purchasing, contractor materials, leased equipment and wholesale. To get a household bill we need the household share of each industry — and Louisville reports by industry, not by customer type, so nobody has published this number. It is the largest judgment in the model.

How the largest share is bounded

General Merchandise is 32.0% of the category and carries more weight than any other single assumption. (That share is computed from the 2025 column. The City’s own printed “% of total” column is calculated off 2024 and reads 29.7%, so we recomputed rather than reprint a figure that lags a year.) Two constraints bracket it, and they are not equally strong:

• The ceiling, which is firm. A household cannot buy more taxable delivered goods than it spends in total. Working from an $89,972 total expenditure basket for this income band, most of it untaxed in Colorado, we put the ceiling on taxable delivered goods near $16,500 a year, which binds the share at 50.4%. This ceiling is reasoned, not measured — it is our judgment about how much of that basket could be taxable delivered goods, not a published statistic. It is the firmest constraint available and it is also the place a reader who wants to push on this analysis should push first.
• The floor, which is not. US household e-commerce of roughly $8,300 a year must fit inside household retail purchasing. This binds at 27.8%.

The share is not merely unpublished. We asked the City of Louisville directly whether its records distinguish deliveries to residential from commercial addresses, and they do not: a sales tax return identifies the jurisdiction a sale was sourced to, not the character of the buyer, and only an audit of an individual vendor's books reveals who bought what. Bounding is therefore not a stand-in for a measurement we expect to obtain later. It is the only method available — to us, or to anyone.

An earlier reading of this bound concluded the share must be at least 70%, which would put the cost near $75 a month. That reading tested household e-commerce against the General Merchandise line alone, and it was a category error: e-commerce is a sales channel, not an industry. Against the industries household e-commerce actually lands in, the bound clears comfortably.

We asked the City how it classifies, expecting one of two answers — that all electronic retail books to General Merchandise, or that it is distributed by vendor sector. The answer is neither. Businesses that register with the City directly are classified from what they themselves report. Payments arriving through the state's Sales and Use Tax System, the portal remote sellers use to file into self-collecting cities and therefore the portal much of this category files through, default to General Merchandise.

That makes General Merchandise partly a residual rather than an industry, and it cuts both ways. The default does place remote consumer retail in that line, which is what the 70% reading assumed. By the same mechanism it places remote business supply there too: an out-of-state industrial supplier filing through the portal is indistinguishable from an online retailer in that column. The two effects offset, and the line's household content tends toward the mix of remote sellers generally rather than toward a distinctively consumer figure.

We then asked how much of the category arrives that way, and the City answered on 21 September 2026: about two-thirds. That bounds the residual instead of leaving it open. Two-thirds of the base arrives through the portal while General Merchandise is 32.0% of it — so the default bucket is half the size of the channel feeding it, and at least half of what files through the portal is sitting in lines the City did classify. The default is real, and it is not what most of this category is. What it does not do is split the line: the household share within General Merchandise is still unmeasured, and the bounds above remain the answer.

We publish 48% — just under the ceiling. Roughly half of this base is business purchasing, so 48% puts General Merchandise a shade more household than the category it sits in, which is the direction that overstates what a household pays. The bottom of the permitted span would require believing that fewer than three dollars in ten of outside-city General Merchandise is bought by a household, and that is a strong claim with nothing behind it but arithmetic permission.

Groceries are in Louisville's base and should not be in ours

Colorado exempts food for domestic home consumption, and in state-collected jurisdictions that exemption flows through automatically. A newly incorporated statutory town is state-collected, and Niwot's measure exempts groceries by its own terms rather than by later discretion. Louisville, as a home-rule self-collecting city, is not bound by the state exemption and taxes groceries.

Building Niwot's household spending from Louisville's industry table therefore imports a line Niwot would not levy. We remove 72.5% of it — not all, because part of a grocery vendor's sales stays taxable even in Niwot (candy, soft drinks, prepared food). This is worth about $1.20 a month, and until 14 September 2026 the published figure included it in full.

An independent check from federal spending data

Everything above builds household spending from what a neighbouring city collects. A different route goes through what households report buying. The Bureau of Labor Statistics’ Consumer Expenditure Survey for 2023, adjusted for the West’s prices and Niwot’s income, puts a typical household’s spending on goods and services Colorado taxes at about $19,000 a year, wherever it is spent. For a household the size of a Niwot condominium (1.1 people) it is about $9,700. Our figures, $19,150 for the average household and $10,642 for a condominium, land within 10% of those.

Two caveats limit what this shows. The survey is known to undercount spending against national totals, and it cannot say how much Niwot residents buy in Niwot or by delivery rather than in Longmont or Boulder, which Niwot would not tax. So it confirms the size of our figure without proving which way it errs. Reproduce it with working/cost-per-household/condo_budget.py.

Junction 4 — Spending in Niwot

What the 28% actually claims

A percentage is impossible to check. Restated, 28% means $4,032 a year, or $336 a month, per household, spent at Niwot businesses. Only 33 of Niwot's 92 open businesses generate taxable retail sales — 17 retail and gift shops and 16 restaurants and cafes — because Colorado does not tax most services. There is no grocery anchor and no large-format store to absorb that spending.

Federal expenditure survey data puts food away from home at $7,652 a year for the highest income quintile, which is the right band for Niwot. So the 28% requires:

So the real assumption inside the 28% is that a Niwot household spends two of every five restaurant dollars in Niwot. That is checkable by any resident against their own habits, which is the point of restating it. The risk is one-sided: if the true capture is lower, residents are a smaller share of the base, which lowers household cost.

Why the rate here is 1.5% and not 2.5%

Downtown, the municipal tax replaces the existing 1% district tax, so the net increase is 1.5 percentage points rather than 2.5. Earlier versions charged the full 2.5% on all local spending and described that as conservative on the grounds that no household knows what share of its local spending falls inside the district. That reasoning is true of households but not of this number: the local line is derived as a share of the district's own base, so it is in-district by construction.

The honest accounting has two halves, and publishing one without the other would mislead. Charging 2.5% overstated. But the district-derived line also omits household spending at Niwot businesses outside the district entirely, which understates. Those partly offset; break-even is at $1,626 a year of off-district spending. Most Niwot taxable retail is inside the district, so we estimate $400 and include it explicitly.

Junction 5 — From an Average Household to a Median Home

Everything above produces the spending of an average Niwot household. The published figure is for a median-value home. These are not the same, and the property side and the spending side must be indexed consistently.

The spending side is indexed once, to the median-value home, by the constant above. Everything downstream of it — every percentile, every neighbourhood — then carries one of two spending lines, according to whether the dwelling is a house ($473 a year) or a condominium or townhome ($244). The two are the canonical figure scaled by measured household size, and the scaling is mean-preserving over all 1,735 dwellings, so no revenue figure moves. The published range widens across dwellings because the property levy varies and because a condominium household is measurably smaller — not because spending varies with home value.

Within a dwelling type, spending is still held flat, and that is wrong in direction. A household in a $3.3 million house does not buy the same quantity of taxable goods as one in a $685,000 house, and both carry $473 here. We have not measured how much more, so rather than apply an uncited elasticity we hold the line flat within a type and state which way it errs: it overstates the least valuable houses and understates the most valuable ones. At the median it makes no difference at all, by construction, so the headline figure is unaffected either way.

Scaling by home value was considered and rejected. Raising value to the power 0.65 would have published a headline of $58 and a middle-80% band of $27 to $111. It was declined on 16 September 2026 for two reasons. The first is that 0.65 is an income elasticity applied to a home value, and it is labelled UNSOURCED where it is defined; home value is a stock and taxable spending is a flow, and Niwot has many long-tenured owners whose houses have appreciated faster than their spending. The second is that it moved only the three neighbourhood comparisons that were genuinely close, and moved all three in our own favour, while changing nothing else — which is a method chosen for its answer. Household size was adopted instead because its scaler is measured twice, by different people for different reasons: ACS household size by units in structure, and registered voters per dwelling, which agree that a condominium household is 0.48 of the town mean. It also costs us two of those three comparisons. Closing the within-type question properly needs the elasticity of taxable spending to income, which the Consumer Expenditure Survey reports by income decile; it is listed under Open Questions for that reason.

One trap deserves naming, because it is easy to fall into and it would move the figure by several dollars. Niwot's income distribution is considerably more skewed (mean 1.34 times median) than its home value distribution (1.17). It is tempting to conclude that the mean-to-median correction should therefore be much larger. It should not. The published figure is indexed on the median-value home, not the median-income household. The income-skew correction applies to the latter. Applying it here would double-count.

The Full Derivation

The vehicle and remodel lines are real costs but they are not monthly bills, so they are disclosed separately rather than blended into the headline. A car purchase every seven years is not the same kind of number as a utility bill, and averaging it into one makes the figure less clear, not more honest.

Revised 9 October 2026. The vehicle line used a $35,000 car, which put the all-in figure at $77 a month. The budget’s own vehicle use tax works out to about $150 per household a year, the same as a $42,000 car every seven years, so the example now matches the budget and the all-in figure is $79. The $63 recurring figure does not include vehicles and did not change.

Distribution Across Dwellings

Bookends worth disclosing rather than hiding: condominiums and townhomes run $26 to $29 a month, and houses above $2.5 million run over $94.

The same five dwellings under the other method. The table above holds the spending line flat within a dwelling type, so every house from the 25th percentile to the 90th carries the same $473. Scaling spending by home value instead — the method used in the committee’s neighbourhood letters, and set out under The choice of method is itself an uncertainty — spreads it:

Both columns sum to the same town-wide total — each is mean-preserving over all 1,735 dwellings — so no revenue figure anywhere depends on which is used. The second column is steeper at both ends: it costs the cheapest dwellings more and the dearest dwellings considerably more. The published figures are the flatter of the two, which raises the floor and lowers the ceiling relative to the alternative.

Every range on this page varies one thing: the dwelling. Spending is held at the central estimate throughout, and takes one of the two values above according to the dwelling’s type. That is worth stating because it was not true until 14 September 2026. Ranges built over home value had been taking their floor from a lower spending estimate and their ceiling from the central one — switching assumptions at one end and not the other. The effect was that the published floor sat about $3 below what this model gives for the same dwelling, and the width of the range was part household variation and part our own uncertainty, with no way for a reader to tell which was which.

These are different things and they should not share an interval. How much your neighbour pays differs from you because their house is worth more: that is a fact about Niwot. How confident we are in the spending estimate is a fact about us. Blending them produces a range that looks like it describes households while half its width describes our doubt. The spending estimate spans about $3.50 a month from its low case to the published one; it applies to every household equally, and it is disclosed here rather than folded into the ranges above.

Every figure in this table is before the Senior and Disabled Veteran Homestead Exemption (CRS 39-3-203), which exempts 50% of the first $200,000 of actual value. A qualifying household in the median home pays about $27 a year less than the figure above — roughly $2 a month. The published figure does not take it, because most households cannot claim it; the tax calculator does, for those who can. Disclosed here because the omission runs against the household, and because the same exemption is worth far more under a property-tax-only improvement district — about $116 a year at 17 mills and $224 at the typical neighborhood district’s 33 — which is a real difference between the two options for a senior.

How Uncertain This Is, and Where the Uncertainty Lives

Drawing every uncertain input 200,000 times across its stated range gives:

The published $62.66 sits at the 52nd percentile, which is what a central estimate should look like. Standard deviation is $4.97.

The variance decomposes as follows. This is the single most useful table in this document for anyone deciding where to spend effort:

Nearly four-fifths of the uncertainty is a single thing, and it is not any of the judgments this document spends most of its length defending. It is the Census margin of error on Niwot's own household income. Niwot is a small place, the ACS sample there is thin, and the resulting margins are wide. If that one input were known exactly, the standard deviation would fall from $4.97 to $1.82 and every other question in this document would become a rounding detail.

This is worth stating plainly because it cuts against the natural instinct. The arguments a critic is most likely to raise — the household share, the 28%, the treatment of groceries — are between them under 8% of the uncertainty.

A caveat on that table

The Census margins are published figures. Every other range in the simulation is our judgment about how wrong an input could plausibly be. The decomposition should be read as a guide to where effort is worth spending, not as an interval anyone else would reproduce from the same data.

The choice of method is itself an uncertainty

Everything above holds one thing fixed: that spending is spread across dwellings by household size. That is a choice, not a measurement, and the simulation does not vary it. The nine inputs it does vary are all quantities inside a method that is assumed correct, so the $4.97 standard deviation is a conditional figure — conditional on the method. Reporting it without saying so would understate how uncertain this is, and by the largest single term.

There are two defensible methods. Both spread the same town-wide total, so neither moves any revenue figure, the pro forma or the ballot-issue estimates; they differ only in how they distribute it.

• Household size. A condominium household is 0.48 of the town mean. Two independent measures agree on that ratio — Census structure type and the county voter roll — which is why it is used for the figures on this site and in the filed ballot summaries. It is a measurement, but it is silent within a dwelling type: it gives the same spending line to a $590,000 house and a $3.3 million one.
• Home value. Spending scales as value to the power 0.549, the income elasticity of taxable spending fitted to the Consumer Expenditure Survey’s five income quintiles for 2023 (R² = 0.994), restricted to the categories Colorado actually taxes. It answers the within-type question the other method cannot, and it agrees with the measurement where the measurement exists — on the two condominium associations, where occupancy was observed, it lands within 7%. It is used for the per-neighbourhood figures in the committee’s 2026 letters. Its weakness is stated plainly under Open Questions: it assumes income tracks home value, and that link cannot be measured in Niwot.

Running the same 200,000 draws with the method treated as unknown rather than fixed:

The width barely moves — $4.97 to $5.14 — because the two methods disagree about the centre far more than about the spread. What moves is where the published figure sits in it. Against the conditional interval, $62.66 is at the 52nd percentile, a coin flip. Against the unconditional one it is at the 67th, and the probability that the true cost exceeds it falls from about 48% to 33%.

This is why the published figure is $63 and not $59. Applied to the median home, the home-value method gives $58.67 a month rather than $62.66. We publish the higher one. The reason is not that the lower one is wrong — it is that when two defensible methods disagree about the cost of something a household is being asked to vote for, the honest choice is the one that does not understate it, and stating both is better than quietly averaging them. A reader who prefers the other method should read $59, and should know that the committee’s own letters use it.

The rule that sets the published range

Two methods that disagree cannot both set a published figure, and choosing whichever is convenient at each end is the basis-switching defect corrected on 14 September and recorded below. So the rule is stated once and applied at every point:

Where the two methods agree within the model’s own noise, publish the central estimate. Where they genuinely disagree, publish the one that does not understate the cost.

“Within the noise” is measured rather than asserted. The standard deviation below is the 200,000-draw simulation evaluated at that dwelling:

The floor and the typical home are cases where the two methods are inside the noise, so the central estimate stands — and at the 10th percentile the simulation’s own median is $28.34, which rounds to the published $28 rather than to the home-value method’s $30. The published floor is the better estimate, not a legacy one.

The ceiling is the one place the rule moves anything, and it moves it up. There the two methods are 2.7 standard deviations apart, which is a real disagreement and not rounding, and the published figure is therefore the higher: a household in the top decile of cost may pay about $108 a month. Publishing $94 would have been the lower ceiling — the flattering direction, not the cautious one. This is the direction to be wrong in for a cost a household is being asked to vote for, and it is why the rule exists rather than a preference for whichever basis was convenient.

The TABOR fiscal summaries were filed on 18 September 2026 stating $28 to $94, before this rule existed. That filing is frozen and correct as filed; the site’s ceiling is now $108, and the divergence is deliberate and stated here rather than left for someone else to find. Reproduce the test with python3 working/cost-per-household/distribution.py --rule.

What Would Have to Be True for This to Be Wrong

For each input, how far it would have to move on its own — everything else held at its best estimate — to reach a given figure:

No single input moves the answer to $65 without taking a value that is either arithmetically impossible or outside its own published margin, with one exception: the income elasticity. That is the honest summary of the model's exposure, and the direction of that exposure runs against the published figure, not for it.

Who Funds the Town

The same model read from the other side answers where the town's money comes from. Four categories, because three do not close: about 11% of revenue is state gas-tax sharing, county road money and lottery-funded Conservation Trust proceeds that nobody in Niwot pays.

The property component is observed rather than modelled: 1,742 residential parcels against 65 commercial, from assessor data, producing 94% of the pro forma's own property tax line. Commercial property is assessed at 25% of market value against a home's 6.8%, which is why 65 parcels carry 27% of the levy. Both counts moved in September 2026, when the classification was audited. A defect in the rule order had booked 18 houses on Monarch Park Court and Monarch Road as commercial, because they share the Monarch Park PUD subdivision name with the business park and the subdivision test was running ahead of the county's own residential account type rather than behind it. A second pass then found three addresses whose rows were duplicates of one parcel, and two addresses that carry several different parcels which the street-address join had collapsed onto one owner — among them 33 acres of Boulder County open space under a conservation easement, which had been sitting inside the residential levy. The earlier figures were 1,729 residential, 84 commercial and 30% of the levy.

The business row is not a downtown row. Reading "Niwot businesses" as the shops on Second Avenue inverts where the money is. The $206.7 million of commercial market value inside the proposed boundary divides like this:

Boulder Tech Center alone is worth more than every other commercial property in Niwot put together, and with Monarch Park the two business parks are 72% of the base the commercial levy falls on — against 18% for Second Avenue. Because commercial property is assessed at a flat 25%, those value shares are also the levy shares. The purchasing side leans the same way: within the business bucket of the sales base, the equipment component is anchored to $13.1 million of business personal property — an actual value, spanning both business parks — rather than to retail fixtures downtown, while the supplies and telecom components are spread across all 190 businesses. Every parcel in both parks tests inside the proposed boundary.

Two naming traps are worth stating plainly, because each has already produced an error in this table. "Monarch Park" names two different things: the business park on Monarch Park Place and Horizon Lane, and a subdivision of houses on Monarch Park Court and Monarch Road that carries the same recorded PUD name. Only the first is commercial, and reading the recorded name alone put 18 houses in the commercial column. And there is no Peppertree Business Park. A single parcel at 6870 Peppertree was carried here as one until 14 September 2026; it is Outlot B of Peppertree Estates, 2.25 acres owned by Boulder County Parks & Open Space. The assessor lists the address as "6870 PEPPERTREE **", and the trailing marker broke the join that would have shown the county as owner, so the blank was filled in by hand with a guess.

The resident share is not a point estimate. Across the open range of the household share it runs from 47% to 54%, and it should be quoted as a range. In prose we round it to about 49% rather than carrying a decimal, because the correction of 30 September leaves about half a point unfixed in the same direction — see that revision entry. Three superseded figures should not be cited: 45%, built on the retention error described above; 48.3%, which rested on an underived household split of the use tax line; and 50.1%, which rested on a household purchasing figure that scaled by the wrong income ratio and counted groceries this town would not tax. 48.8% is not a return to 48.3%. They are different corrections to different inputs that happen to land half a point apart, and 48.3% remains superseded.

The visitor band, and what supports it

The visitor figure rests on the least evidence of the four. It depends on the same 28% discussed above, which is built from estimated household spending and has never been observed at the register. Here is what does support it.

What supports it is the district's own history. Between 2019 and 2022 downtown taxable sales rose $6.0 million in real terms, a 23% increase, then gave all of it back by 2025, ending slightly below where they started. For residents to have produced that surge, every one of 1,735 households would have had to add roughly $290 a month of local spending on top of the $339 the 28% already credits them with, and sustain it for two years while restaurants were capacity-restricted. The reversion matters as much as the surge: a permanent shift of resident spending toward local businesses leaves a residue, whereas a transient flow of day-trip visitors receding as travel normalised looks exactly like this.

What it establishes is that the visitor channel is large and real. It does not pin the number at 28%.

Reconciling Burden and Incidence

A reader who takes the town's revenue, identifies the residents' share, and divides by 1,735 households will not get $63, and should not expect to. At the canonical household share, residents fund about 49% of the $4,598,000 the pro forma models for 2030 — roughly $2.25 million, or $1,295 per dwelling per year. That is not the same quantity as the $752 this document publishes, and the whole of the difference is convention rather than disagreement about a rate.

Five conventions separate them, and a residual remains after all five. The bridge is reproducible from working/cost-per-household/reconcile.py:

Figures are rounded to the nearest dollar, so a column may be a dollar out.

• Line coverage — $279. The headline is property and sales tax only. The revenue side also counts use tax on vehicle purchases (about $116 a year spread across all homes, disclosed here as a one-time cost instead), roughly $72 of building permit fees, $28 of franchise and minor municipal fees, $17 of specific ownership tax and $10 of marijuana tax. These are per household in 2025 dollars, the same basis as the headline — the pro forma’s 2030 column states the same money in nominal 2030 dollars, about 16% higher. Most of that is not new money. Specific ownership tax is already collected at annual vehicle registration and split among taxing districts — incorporation makes the town one of the recipients rather than adding a charge. Building permits are already paid to Boulder County today, and under the modelled contract 85% of the fee passes straight through to the contractor performing the inspection. What is genuinely new is the vehicle use tax and the marijuana tax.
• Mean against median — $59. A per-household division gives the mean home, $1,276,872, against the median $1,095,000. Niwot's home values are skewed upward by a long tail, so the mean describes no actual house. The median is the better answer to “what will this cost me.”
• Dollar year — $128. The pro forma's 2030 column is in 2030 nominal dollars; this document is in 2025 dollars, because that is what a reader can check against their own assessment notice. Five years of 3% inflation is a 16% difference on its own.
• Lumpy items — $42. Remodel materials are real revenue to the town but are not a monthly bill, so they sit with the vehicle tax in the all-in figure rather than in the headline.
• Spending basis — $35. Two models sit behind this document. louisville-benchmark/model.py lets a household's spending track its home value, which at the median gives $59.70 a month; cost-per-household/distribution.py, which produces the published figures, holds spending at the canonical share and scales it by measured household size instead, giving $62.66. The $2.96 a month between them is a documented convention, not drift, and the published figures use the second. It is named here rather than folded into the residual: netting the two would report a $9 credit that is really a $26 disagreement and a $35 convention pulling opposite ways, and a reader could not take either apart.

The residual, $26, is disclosed rather than absorbed. After the four conventions above it, the revenue-side reconstruction still lands about $26 a year above the household-side derivation — roughly 4%, or about $2.19 a month. It was $55 until 30 September 2026; half of what had been called unexplained turned out to be two defects in the revenue-side numerator, and correcting them is what moved it. The two routes are built from different data (the pro forma's modelled property line against parcel-level assessor values), so exact agreement would be surprising rather than reassuring. We publish the household-side figure, which is the lower of the two, and show the gap instead of splitting the difference. A reader who prefers the revenue-side route should read the published figure as about $2.19 a month conservative; that is the honest size of the disagreement, and it runs in the direction of overstating what a household pays.

Residents fund roughly half the town's revenue; businesses about 29%, visitors about 11%, and state and formula money the remainder. The reason a large base and a modest household bill are both true is that households are not the whole base.

Comparability Cautions

Two published benchmarks, on different bases. The pro forma benchmarks against both Louisville and Lyons, and both figures are verified against each town’s own published documents: Louisville’s Finance Committee sales tax report, and the finance updates in Lyons’ Board of Trustees packets, which cover the full 2025 year. They are not the same instrument. Louisville is home rule and self-collecting, and its “Outside City” category covers every vendor beyond its boundary. Lyons is a statutory town whose sales tax the Colorado Department of Revenue collects, and its non-local category is the Town’s own split of that revenue, including utilities. So Lyons corroborates the magnitude rather than reconciling category for category, and where Lyons is used above to test the income elasticity, that caveat applies in full.

The $16,500 ceiling is reasoned, not measured. Repeated here because it carries more weight than any other single judgment: it is our estimate of how much of an $89,972 expenditure basket could be taxable delivered goods, and it is not a published statistic.

The local spending share is an assumption. The 28% moves the monthly figure by roughly $3.60 across its plausible range and nothing observes it.

This is a planning model, not an adopted budget. The pro forma it draws on is not an adopted budget either. An elected Town Council would set a budget annually, and would do so with actual collections in hand rather than these estimates.

Revision History

This figure has moved three times in September 2026, and its ranges were rebased twice. A later entry restates who pays without moving the figure at all. The reasons should be on the record rather than inferred.

The 28 September entry restates uncertainty rather than any figure, and the 16 September entry is the only other one of these that is not a correction. Nothing was found to be wrong; a convention that had always been an approximation was replaced by a measurement, and the approximation is why the band was narrow. Holding spending flat across every dwelling and publishing a $44 floor were the same choice seen twice.

Of the earlier entries: the figure returned to where it started, and it would be reasonable to ask whether anything was learned. The reasoning is entirely different: the 12 September value was correct given what was then known, and the 14 September corrections are findings that postdate it. Every correction made in September moved the figure in the same direction — downward — except the first.

The rebasing the same day is a different kind of entry, and worth separating from the four corrections above it. Nothing was recomputed and the headline did not move; what changed is that the ranges stopped answering two questions at once. It was raised by the niwottogether.org campaign, which had reproduced this paper's distribution table exactly and could not reproduce the $41 floor from it — the floor came from a spending assumption the table does not use. Their diagnosis was that the figure was underivable; it was derivable, but only by switching an assumption mid-range, which is the more useful finding and the one acted on here.

Open Questions

One of the four below has been answered since this paper was last revised, and is marked so rather than deleted — a methodology paper that quietly drops its open questions is not one a reader can audit.

Three questions that stood here were put to the City of Louisville's Finance Department and answered on 14 and 21 September 2026, and all three answers are recorded above. The residential and commercial split within Outside City is not something the City records, and could not be produced without auditing each vendor individually. The classification question, and the share of the category arriving through the state portal, are answered in the paragraphs on General Merchandise. The household share is not going to be measured by anyone; the bounds are the answer, and we are grateful to the City for saying so plainly.

1. The 33 downtown businesses: what share of customers come from outside Niwot? The only question here that needs nobody else's cooperation.
2. Niwot household income at the boundary rather than the census place. The largest single uncertainty in the model. Census block-group data aggregated to the proposed boundary would address both the footprint gap and the margin.
3. The elasticity of taxable spending to home value. Answered, 23 September 2026. This question asked for a fitted elasticity in place of a judgment, from the Consumer Expenditure Survey, with its categories classified as taxable or exempt in Colorado. That is what was done: restricting the 2023 survey to the categories Colorado taxes — groceries, shelter, utilities and cash contributions excluded — and fitting across the five income quintiles gives 0.549, R² = 0.994, with the taxable share of total expenditure holding between 31% and 35% across all five. It replaced an exponent of 0.65 that was an elasticity of general consumption and carried no citation. The committee’s neighbourhood letters are computed on it.
   
   What remains open is the assumption underneath it, and it cannot be closed here. The elasticity is to income; applying it to home value assumes income tracks value. Census table B25121 tops out at a home-value bracket of “$500,000 or more” and 93% of Niwot’s owner-occupied homes are above it — checked at both place and county level, where the margins are clean and the ceiling is identical. The limit is in how the table is published, not in our sample, so no choice of geography fixes it. In Niwot the assumption probably overstates the gradient, because long-tenured owners sit in appreciated homes on modest incomes. The defensible sentence, and the one to use, is: spending scales with home value at 0.55, the income elasticity of taxable spending measured from the 2023 Consumer Expenditure Survey, on the assumption that income scales with home value. The first half is sourced; the second is an assumption stated rather than buried.
4. The income elasticity used in the Louisville transfer. A multi-town regression of taxable base per household on income would replace a judgment with a fitted value and a standard error. Partly addressed: the 0.549 above is a fitted elasticity of taxable spending to income, and it is the same quantity this transfer needs. Substituting it for the 0.65 used here would lower the published figure by about $0.25 a month, so the published exponent is the more cautious of the two and has been left in place. What a multi-town regression would still add is a standard error, which neither value carries.

Reproducing This

Every figure in this document is produced by scripts in the campaign's working repository, and the arithmetic there is the authority: where a narrative disagrees with a script, the script is correct.

Sources

• City of Louisville, Colorado. Finance Committee packet, 15 January 2026, Revenue and Sales/Use Tax Reports, November 2025, pages 39–45. Provided to the committee directly by the City. Cited excerpt.
• US Census Bureau, American Community Survey 2024 5-year estimates. Tables B19025 (aggregate household income), B11001 (households), B19301 (per capita income), B19013 (median household income). Niwot CDP, Louisville city, Lyons town, Colorado.
• Colorado Department of Revenue, DR 0100 net taxable sales by city and month, published as open data at data.colorado.gov (resource 2yhn-3dbj).
• Boulder County Assessor, 2025 actual values, parcel level, for the 1,735 dwellings inside the proposed boundary.
• Boulder County, Niwot Local Improvement District receipts, 2025 actual.
• US Bureau of Labor Statistics, Consumer Expenditure Survey 2024.
• US Bureau of Labor Statistics, Consumer Expenditure Survey 2023, by size of consumer unit and by region of residence, via the BLS public data API.
• Colorado Revised Statutes 39-26-707 (food for domestic home consumption) and 39-3-203 (senior and disabled veteran homestead exemption).
• Committee demographer, proposed boundary population, 4,305.