Estimating Cannabinoids Sold from Marijuana Sales Data in Washington State Watch Me Pull a Number out of My Hat

Estimating cannabinoids sold from marijuana sales data in Washington state Watch me pull a number out of my hat

Jason Williams1 with assistance from Caleb Banta-Green1 & Dan Burgard2

1Alcohol & Drug Abuse Institute University of Washington 2Chemistry Department University of Puget Sound

20 March 2018 / ADAI Brown Bag

Jason Williams (ADAI) Measuring THC 1 / 42 Acknowledgments

Thanks to: The Washington State Dedicated Marijuana Fund for research at the University of Washington for funding support Looking Glass Analytics for ﬂattening the database

Jason Williams (ADAI) Measuring THC 2 / 42 Today’s outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 3 / 42 Outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 4 / 42 Why measure cannabinoids? Important questions

Policy analysis To what extent does the THC sold via licensed stores account for the THC metabolite found in wastewater, implying displacement of the illegal market? Marketing studies To what extent is the market in Washington moving to higher potency products? Microeconomics What is the price elasticity of THC potency? Public health How much THC is out there, and where?

Jason Williams (ADAI) Measuring THC 5 / 42 This should be easy We have the data...

Washington state Liquor and Cannabis Board (LCB) mandates use of a seed-to-sale database to Track every plant Follow each plant as it is reﬁned and turned into products Track testing results for every product All to demonstrate products are not being diverted. Also, to facilitate excise tax collection, etc.

Jason Williams (ADAI) Measuring THC 6 / 42 This should be easy We have the data...

Jason Williams (ADAI) Measuring THC 6 / 42 Outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 7 / 42 How it should work Alcohol example

Item Packages Sold Package Count ABV% Volume Per Count (L) Alcohol (L) Goodstuff Vodka 1 1 40 1.75 0.7 Sawbuck Red 2 1 15.2 .75 0.228 Lily White 1 1 13.8 .75 0.1035 Crusher Lager 1 6 5.2 .355 0.11076

Jason Williams (ADAI) Measuring THC 8 / 42 How it should work Alcohol example

Total liters of pure alcohol sold = L of Alcohol For each product i, L of Product × L of Product = Pn ABVi i=1 Packages Soldi × Package Counti × Volume Per Counti × 100

Jason Williams (ADAI) Measuring THC 9 / 42 How it should work THC example

With marijuana, you have to: Account for different measures of size: Grams Servings per container, or package count Volume Match potency unit of measure to product measure Percent by weight mg per serving Percent by volume

Jason Williams (ADAI) Measuring THC 10 / 42 How it should work THC example

Determined by: Manufacturer Lab LCB Store Item Package Count (L THC Potency Packages/ THC (mg) for liquids) Potency UOM Grams Sold HazyNut Brownies 1 5.05 mg/unit 2 10.1 HazyNut Brownies 10 5.05 mg/unit 3 151.5 VapoRide Concentrate 0.1 75 mg/L 3 22.5 Hop No Hop Infusion 0.25 10 mg/L 2 5.0 Jumbo Java K-Cups 6 4.85 mg/unit 1 29.1 Slow Dive Sativa 1 24.56 % by 0.5 122.8 weight Emotion Lotion 5oz tube 0.148 300 mg/L 1 44.4

UOM: unit of measure

Jason Williams (ADAI) Measuring THC 11 / 42 Outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 12 / 42 So what’s the problem? Solid edibles

PackagesSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 5 4.58 4.36 4.43 0.24 Oakor 1:1 Slips 1:1 Slips

Jason Williams (ADAI) Measuring THC 13 / 42 So what’s the problem? Solid edibles

PackagesSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 5 4.58 4.36 4.43 0.24 Oakor 1:1 Slips 1:1 Slips Assuming this is 4.58mg of THC per slip, did this sale amount to 22.9mg, or 458?

Jason Williams (ADAI) Measuring THC 14 / 42 So what’s the problem? Solid edibles

PackagesSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 1 0.006 0.06 0.05 0 Catapult Decaf Individual Catapult Decaf Individual 1 0.11 0.06 0.05 0 Catapult Decaf Individual Catapult Decaf Individual 1 0.9 0.9 0.85 0 RCC010D Dark Choc CBD 3- RCC010D Dark Choc CBD 3- Pack Pack 1 0.9 0.9 0.85 0 EH 4.20 Bar 3pk CBD Dark RCC010D Dark Choc CBD 3- Choc. 30mg Pack 1 0.91 0.91 0 0 RCZ010 Dark Choc Hazelnut RCZ010 Dark Choc Hazelnut 2 0.92 0.91 0.01 0 Peanut Butter Cup Peanut Butter Cup

Jason Williams (ADAI) Measuring THC 15 / 42 string87

10.37 mg / g Accreditation No. 73653 ISO / IEC 17025:2005 ACCREDITATED COMPANY Certificate No. L12-227 Medical Cannabis A8599C

CLIENT GFarmaLabs STRAIN Blue Diamond SAMPLE TYPE Flower DATE OF ANALYSIS 2/28/2017 CSA LAB ID NO. D06250687 RECEIVED ON 2/24/2017 SoSAMPLE what’s ID NO. the problem?NA EXPIRATION DATE 3/29/2017 LOT NO. NA TEST METHODS HPLC-002 MOISTURE CONTENT 9.5 % WATER ACTIVITY 0.371 CANNABINOID CONCENTRATION (mg/g) PERCENT by MASS Delta 9 Tetrahydrocannabinol D9 THC 10.37 1.04% Delta 9 Tetrahydrocannabinol A D9 THCA 220.22 22.02% Cannabidiol CBD Tetrahydrocannabivarin THCV

Total Cannabinoids

23.65 %

Jason Williams (ADAI) Measuring THC 16 / 42

Yeast & Mold Count NOT REQUESTED Aerobic Bacteria Count NOT REQUESTED Total Coliform Enumeration NOT REQUESTED Bile Tolerant Gram - Negative Bacteria NOT REQUESTED E. coli NOT REQUESTED Salmonella NOT REQUESTED So what’s the problem? Solid edibles

PackagesSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 1 0 5.12 0 0 BLACKBERRY chocolate 10HBB - HYBRID Blackberry Choc 5mg x 2/pack (10mg) 1 0 0 0 0 FW Hot Cocoa Mocha Catapult Hot Cocoa Java Sin- gle Serving 1 0.03 0.03 0 0 FW Hot Chocolate Catapult Hot Cocoa Java Sin- gle Serving 2 0 8.4 0 0 Combo_TRIPLE_30mg Combo_TRIPLE_30mg 2 0.2 0.2 2.16 0 Capsule CBD 10:1- 10mg (10 Capsule CBD 10:1- 10mg (10 Serve) Serve) 1 0 9.41 9.41 0 Tomato Basil Soup 1 Box Tomato Basil Soup 1 Box

Jason Williams (ADAI) Measuring THC 17 / 42 So what’s the problem? Liquid edibles

PackagesSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 1 5.12 5.12 0.45 0 SM 50mg Dank Drizzle Dank Drizzle 1 5.12 5.12 0.45 0 SM 100mg Dank Drizzle Dank Drizzle 1 0.0112 0.0112 0 0 OC SENSI Northwest Berry Northwest Berry Lemonade, 100mg 16oz, 100mg 4 0.98 0.95 0 0.02 ZootBlast - 30mg Single ZootBlast - 30mg Single 1 0.95 0.95 0 0 ETHOS Oral Spray Create Create(10ml) 1 1.05 1.05 0 0 Tincture Natural Tincture Natural 1 78.7 78.7 3.57 0 ETHOS Water Tincture Jour- Journey(1oz) ney

Jason Williams (ADAI) Measuring THC 18 / 42 So what’s the problem? Liquid edibles

PackagesSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 1 5.12 5.12 0.45 0 SM 50mg Dank Drizzle Dank Drizzle 1 5.12 5.12 0.45 0 SM 100mg Dank Drizzle Dank Drizzle 1 53.9 53.9 95.9 0 PTNZ - Dropperz - CBD PTNZ - Dropperz - CBD Honey Lemon 100mg Honey Lemon 100mg 1 23.7 23.7 28.3 0 FAIRHAVEN RCBD1-60 FAIRHAVEN RCBD1-60 12.19 12.19 1 54.19 54.19 0 0 Strawberry Lemonade, 12 oz, Strawberry Lemonade, 12 oz, 60 mg 60 mg 1 59.94 59.94 0 0 Strawberry Lemonade, 12oz, Strawberry Lemonade, 12oz, 80mg 80mg 1 54.4 54.4 69.3 0 BALLARD D27-CBD 5.11 BALLARD D27-CBD 5.11

Jason Williams (ADAI) Measuring THC 19 / 42 So what’s the problem? Concentrates

GramsSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 1 1.4 0.2 0 1.4 Royal THC Vape Pink Pas- Royal THC Vape Pink Pas- sion sion 1 1.4 0.2 0 1.4 Royal THC Vape God Bud Royal THC Vape God Bud 1 39.34 0.61 44.16 0 CCC KIEF 1 Kief 1.155642023 98.4 30.1 68.3 0 Aliens on Moonshine Clear Aliens on Moonshine Clear 1 97.5 51.6 41.9 4.6 White Widow Clear White Widow Clear 1 97.3 96.75 3.6 0.6 GG - Sugar Wax - 1.0 gram GG - Sugar Wax - 1.0 gram package package

Jason Williams (ADAI) Measuring THC 20 / 42 So what’s the problem? Concentrates

Jason Williams (ADAI) Measuring THC 21 / 42 So what’s the problem? Topicals

PackagesSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 2 0 0 0.45 0 Topical Flow Gel Topical Flow Gel 1 0.008 0.08 0 0 Bath Bath 1 0.99 0.53 0.01 0.36 Body Buzz(4oz) Body Buzz(4oz) 1 0.99 0.53 0.01 0.36 Ethos Body Buzz Bath Salt Body Buzz(4oz) 60mg 2 93.1 93.1 0.2 0 PTNZ - Lotionz - (5) Ampules PTNZ - Lotionz - (5) Ampules 300mg 300mg 1 68.8 68.8 0.3 0 PTNZ - Lotionz - (5) Ampules PTNZ - Lotionz - (5) Ampules 300mg 300mg

Jason Williams (ADAI) Measuring THC 22 / 42 So what’s the problem? Mix package

GramsSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 1 0.38 0 6.86 0.43 Harle-Tsu 1g Walden 1g dried ﬂower strain marketed as 20:1 CBD 1 1.4 1.4 4 0 Harlequin 1g Joint Harlequin 1g Joint strain marketed as 5:2 CBD 3.5 2.54 0.11 0.16 2.26 Finished Flower - Medi Haze Finished Flower - Medi Haze strain marketed as between 1:1 and 2:1 CBD

Jason Williams (ADAI) Measuring THC 23 / 42 So what’s the problem? Mix infused

GramsSold TotalTHC THC CBD THCA RetailProductName WholesaleProductName 0.88203125 9.27 1.7 6.73 8.63 Third Kynde - CBD Preroll Third Kynde - CBD Preroll 0.826498423 12.7 12.7 18.7 0 Third Kynde - CBD Preroll Third Kynde - CBD Preroll 1 56.5 10.5 4.2 37.5 Golden Nuggets No. 5 Golden Nuggets No. 5

Jason Williams (ADAI) Measuring THC 24 / 42 So what’s the problem? In sum

1 No pack size. 2 No unit of measurement for potencies required. 3 Not clear whether potencies are per serving or per container. 4 Inexplicable potencies. 5 Idiosyncratic descriptors mean we can’t always pull useful information out of names. 6 No instructions to labs for entering potencies. 7 No data checking.

Jason Williams (ADAI) Measuring THC 25 / 42 So what’s the problem? In sum

Jason Williams (ADAI) Measuring THC 25 / 42 Outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 26 / 42 What do we need? Specifying required parameters

Total milligrams of THC sold = THC For each product i, amount of product × unit of product = Pn i=1 Package Counti × Packages Soldi × Potency Per Counti × Scaling Factori or Pn mg THC Grams Soldi × i=1 g Product i

Jason Williams (ADAI) Measuring THC 27 / 42 What do we need? Specifying required parameters

Pull pack size from names? Get some idea of potency from names? How identify nonsensical potencies versus legitimately low or high values?

Total THCi = THCi + .877 × THCAi

Jason Williams (ADAI) Measuring THC 28 / 42 Outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 29 / 42 Our approach Specifying required parameters

1 Specify allowable high and low values (brackets) for pack size and potency By product type Try to differentiate high CBD products? 2 Set everything outside those brackets to missing. 3 Impute missing data from data we "trust". 4 Vary assumptions to calculate a set of estimates.

Jason Williams (ADAI) Measuring THC 30 / 42 Our approach Specifying required parameters

Jason Williams (ADAI) Measuring THC 30 / 42 Our approach Simple version

Product type: pre-calculation Pack size Packages sold THC potency steps Solid edibles (22) Plug in median for missing PackagesSold Plug in median for missing High 20 12mg Low 1 1mg Liquid edibles (23): Replace Assume all potencies are Use marketing mg per bot- 200mg (nominally 1:1) with for whole bottle, therefore PackagesSold tle. Plug in median for miss- 100mg to represent THC pack size = 1 ing. High 1 100mg Low 1 9mg GramsSold; ignore Sold by weight, therefore Concentrates (24) nominal weight in Plug in median for missing pack size = 1 names High 1 98%bw Low 1 2%bw

Jason Williams (ADAI) Measuring THC 31 / 42 Our approach Simple version

Product type Pack size Packages sold THC potency Assume all potencies are for Topicals (25) IGNOR- whole package, therefore pack PackagesSold Single guess for all = 50mg? ING size = 1 Sold by weight, therefore pack Bud (28) GramsSold Plug in median for missing size = 1 High 1 40%bw Low 1 0.6%bw Sold by weight, therefore pack Mix package (31) GramsSold Plug in median for missing size = 1 High 1 50%bw Low 1 0.6%bw Sold by weight, therefore pack Mix infused (32) GramsSold Plug in median for missing size = 1 High 1 70%bw Low 1 8%bw

Jason Williams (ADAI) Measuring THC 32 / 42 Our approach Simple version

For any sales entry missing one or more parameter, plug in: 1 Lowest possible value 2 Middle possible value (median of those values within brackets) 3 Highest possible value Calculate 3 values of mg THC for each sales entry For complete and trusted sales, low = medium = high Sum → 3 estimated values of THC sold

Jason Williams (ADAI) Measuring THC 33 / 42 Our approach Simple version

Jason Williams (ADAI) Measuring THC 33 / 42 Outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 34 / 42 Results! King County, June 2016, data

Type Rows Bad Pack Potency bracket Bad THC Edibles 99092 41486 41.9% 1mg–12mg 23794 24.0% Liquids 23153 0 9mg–100mg 10025 43.3% Concentrates 119464 0 2%–98% 784 0.7% Bud 684264 0 0.6%–40% 2328 0.3% Mix package 23547 0 0.6%–50% 51 0.2% Mix infused 16858 0 8%–70% 176 1.0% Total 966378 41486 4.3% 37158 3.8%

Jason Williams (ADAI) Measuring THC 35 / 42 Results! King County, June 2016, estimates

Type Pot. bracket Low THC (mg) Mid THC (mg) High THC (mg) Mid % of total Edibles 1mg–12mg 2,682,307 5,298,512 11,850,677 1.4% Liquids 9mg–100mg 985,337 1,333,405 2,007,085 0.3% Conc. 2%–98% 58,890,255 59,202,127 59,323,155 15.4% Bud 0.6%–40% 303,650,863 304,619,042 305,551,430 79.4% Mix package 0.6%–50% 7,761,970 7,786,011 7,824,807 2.0% Mix infusion 8%–70% 5,161,408 5,180,861 5,232,637 1.4% Total 379,132,141 383,419,958 391,789,791 mg difference vs. middle -4,287,817 8,369,833 % difference vs. middle -1.12% 2.18%

Jason Williams (ADAI) Measuring THC 36 / 42 Outline

1 Problems Why would we want to know how much THC or other cannabinoids are being sold? How it should work So what’s the problem?

2 Solutions What do we need? Our approach

3 Results! King County, June 2016 Pierce County, June 2016

Jason Williams (ADAI) Measuring THC 37 / 42 Results! Pierce County, June 2016, data

Type Rows Bad Pack Potency bracket Bad THC Edibles 19240 8768 45.6% 1mg–12mg 4389 22.8% Liquids 5200 0 9mg–100mg 2378 45.7% Concentrates 35148 0 2%–98% 281 0.8% Bud 245934 0 0.6%–40% 665 0.3% Mix package 8008 0 0.6%–50% 0 0.0% Mix infused 3149 0 8%–70% 0 0.0% Total 316679 8768 2.8% 7713 2.4%

Jason Williams (ADAI) Measuring THC 38 / 42 Results! Pierce County, June 2016, estimates

Type Pot. bracket Low THC (mg) Mid THC (mg) High THC (mg) Mid % of total Edibles 1mg–12mg 492,576 780,643 2,410,903 0.5% Liquids 9mg–100mg 202,213 286,750 450,370 0.2% Conc. 2%–98% 21,391,868 21,528,010 21,578,828 14.1% Bud 0.6%–40% 126,625,016 126,976,180 127,303,244 82.9% Mix package 0.6%–50% 2,578,514 2,578,514 2,578,514 1.7% Mix infusion 8%–70% 1,014,812 1,014,812 1,014,812 0.7% Total 152,304,999 153,164,908 155,336,671 mg difference vs. middle -859,909 2,171,763 % difference vs. middle -0.56% 1.42%

Jason Williams (ADAI) Measuring THC 39 / 42 Results! Comparing estimates accounting for population size

Pierce 181.4

182.1 King

176 178 180 182 184 186 188 190 mg per capita

Jason Williams (ADAI) Measuring THC 40 / 42 Summary

The sales data are problematic No ﬁeld for package size No consistent indicator of high CBD product Potency problems With certain assumptions, can still say something about sales Bracket approach still puts trust in values within brackets

Possible improvements Flag high CBD items based on some minimum CBD potency and/or "CBD"? Would miss many "medicinal" products. Conﬁdence intervals via Monte Carlo simulation? High and low bounds < 3.5% wide, so value of "better" CIs unclear.

Jason Williams (ADAI) Measuring THC 41 / 42 Summary

Next step Comparing sales data to consumption estimates from wastewater testing for same time periods and same place For more information anjrw [at] uw [dot] edu

Jason Williams (ADAI) Measuring THC 42 / 42