I accessed a database of METRC lab data for Maryland flower from Oct 2017 (when sales of medical cannabis started) through Dec 2023 (recreational sales started in July 2023).^[1] The original dataset contains data for 407,794 tests. I eliminated tests for:

• Non-flower products
• Stability testing
• R&D testing
• Testing of mg/g, rather than %

Each test contains the following information: 

• Date 
• Strain Name
• Package ID (a unique identifier assigned to the batch of tests conducted on a particular sample on a particular date)
• Lab ID
• Terp % (total terpene content)
• THC % 
• THCA %
• Total Yeast and Mold Counts (TYM)

I aggregated tests by Package ID, that is, combined all tests performed together on each sample, and then dropped Package IDs that did not have all four tests, Terp %, THC %,THCA %, and TYM. That left my final analysis sample containing 83,406 unique Package IDs and 333,624 tests. A particular strain may be provided by multiple suppliers, but supplier IDs were not provided. So analyses can only be performed at the test or strain level, but not at the supplier level. There are a total of 2,914 flower strains used in the analysis. The data that appear in the analysis are all taken directly from the METRC database. I cannot vouch for the accuracy of the lab tests or the strain name desginations. 

4 Tests per PackageID x 83,406 PackageIDs = 333,624 Tests

83,406 PackageIDs comprise 2,914 Unique Strains

The questions I investigated include:

• What do trends in THC % over time look like? I expected to see THC % increase over time.
• If THC % is increasing over time, is the increase due to:

1. Newer strains have higher THC % than old strains, or 
2. THC % in existing strains has been increasing over time (due either to cultivation techniques and/or inaccurate lab testing), or 
3. Some combination of (i) and (ii)?

• What does THC % look like for the most popular strains, that is, the strains with the most tests?
• Are there any other interesting patterns or correlates relating to THC % over time?

THC % by Year

Figure 1 displays annual averages across all tests on Terp %, THCA %, THC %, and Total THC % (= THC % + 0.877 x THCA %). As expected, Total THC % has been increasing over time, reflecting increases in THCA %. There was a big jump in Total THC % between Oct - Dec 2017 and 2018, with decreasing rates of increase in subsequent years. Terp % has also been slightly increasing over time.

Figure 1

Average annual Total THC % is a crude measure, since it doesn’t provide much detail about what’s driving the trend. Figures 2, 3, and 4 provide the annual distributions of test outcomes, which adds a bit of nuance to the information displayed in Figure 1. Figure 2 provides annual distributions of THC % across years. Figure 3 provides the same information as that in Figure 2, but puts each year of data into a separate figure, and Figure 3 provides the distributions in tabular form.

From Figures 2, 3, and 4, we can see three separate interesting trends in the annual distributions :

• Annual distributions are shifting right, and supply of lower THC % products (less than 15%) are disappearing.
• Annual distributions are becoming less variable. That is, the distributions start out being relatively flat and wide (there is a lot of variation in THC % across tests) and end up being tall and thin (there is much less variation in THC % across tests). 
• Over time, there are an increasing number of tests showing THC % > 30% and more than a trivial number of tests showing THC % > 35%. This suggests some high THC % flower is pushing the upper physical limits on THC production by cannabis plants. Clearly, there are also inaccurate lab results included in the database.

Figure 2

Figure 3

Figure 4

Strain Popularity and Longevity

Each lab tests represents a batch of flower sold in the market. So then strains with more tests have greater volumes of market sales. That is, strains with more tests are more popular in the market, either for a given year and/or across years. Popularity may be fleeting (subject to fads) or it may last (experience longevity).

To better understand strain popularity, I aggregated data to the strain level and compared information for strains by the number of years of tests appearing in the dataset for each strain (see Figure 5). So, for example, there were 1,113 strains with only 1 year of testing data. Each of these strains may have more than one set of tests during the year, but the tests were only performed within a single calendar year. At the other extreme, there were 13 strains with tests in each year. As expected, the distribution of the number of strains by number of years of tests follows a power law, that is, most strains have only 1 year of tests, and only a few strains have 7 years of tests.

Figure 5

As seen in Figure 5, the total number of tests performed for each set of strains is relatively similar until you get to strains with tests for all 7 years. However, total tests is not a very useful statistic, because it doesn’t account for the different number of strains in each group or the number of years of tests for each group. To address this problem, I divided the total number of tests per strain and then by the number of years the strain had tests (total tests/strain/year) and compared this statistic for each level of strain longevity. Figure 6 shows that the two types of popularity — popularity within a given year and popularity over time -- are correlated, that is, strains with more years of tests also have more tests per year. So strain popularity tends to last across time.

Figure 6

A slightly different view of the same phenomenon involves comparing the portion of total strains accounted for by each set of strains with the portion of total tests accounted for by each set. Figure 7 shows that while strains with only 1 year of tests account for half of all strains, they account for only 12% of all tests. The 1% of strains with 7 years of tests account for a disproportionate number of tests, 3%, where Figure 6 showed that this is due to the fact that these strains have both more tests each year and over time.

Figure 7

The next step is to disaggregate the high level trend by year to see if the patterns in number of tests per strain have been changing over time. Figure 8 presents this relationship: annual distributions of number of tests per strain per year for strains by longevity.

Figure 8

Figure 8 indicates that: 

• the number of tests per strain (i.e., level of popularity on the market) have not changed much (relatively speaking) for the strains with the least longevity (1 or 2 years in the market) or with the greatest longevity (7 years in the market). The exception to this trend is the most recent year or two — the number of tests (i.e., market popularity) for single year strains increased, while they dropped quite dramatically for 7 year strains. 
• Strains with moderate longevity (3 or 4 years in the market) increased in popularity gradually between 2017 and 2021/2022 then jumped to a relatively stable level during 2021 - 2023. 
• Strains with moderately high longevity (5 - 6 years) increased relatively dramatically from year-to-year between 2017 and 2021, peaked 2021, then dropped just as dramatically from year-to-year between 2021 and 2023. 

The next step is to see how THC % may have affected these patterns in popularity for each of the different sets (by longevity) of strains. These relationships (THC % by strain longevity over time) are displayed in Figure 9. Figure 9 reveals two general relationships:

• Generally speaking strains with greater longevity have had greater THC % than strains with less longevity throughout the 7 year period. This suggests that since at least 2017, the more popular strains, which also have greater longevity, also have higher THC content. Presumably this correlation represents causation: strains are more popular at least in part because they have higher THC %. 
• THC % has been increasing gradually over time for strains with greater longevity, while THC % has been increasing more radically over time for strains with less longevity. This is expected, since strains with greater longevity were starting out with higher levels of THC and thus had less room to grow.

Figure 9

A final analysis in this vein was to take the analysis presented in Figure 9 and examine distributions of THC % for each year separately for strains with, respectively, 1 year of tests, 6 years of tests, and 7 years of tests. This analysis is presented in Figure 10.

Figure 10

New Strain Entrants and Survivors

Given the patterns seen so far, I would guess that new strains entering the market in later years will have higher THC % than strains entering the market in earlier years. And this is exactly what the data show (see Figure 11).

Figure 11

Define a strain as being a survivor if it remains in the market (has tests in the database) for every year after it enters through 2023. So a 2017 survivor has tests for every year from 2017 through 2023 and a 2020 survivor has tests for every year from 2020 through 2023. New strain entrants in 2023 are excluded from the analysis. 

The analysis so far suggests that survivor strains are more likely to have high THC %, and they are more likely to have more tests. Figure 12 shows that this is, indeed, the case. Survivor strains comprise 30% of all strains, but account for 63% of all tests, and they account for 90% of tests yielding Total THC % ³ 30%. 

Figure 12

Figure 12 suggests survivor strains have higher THC % than non-survivor strains. Figure 13 shows that this is, indeed, the case.

Figure 13

So then are the most popular strains also the same as the strains with the highest THC %? Apparently not. Figure 14 (the year of entry for each strain appears in parenthesis after the strain name) shows there’s no overlap between the top 25 strains with the most tests per year and the top 25 strains with the highest average THC %.  Also, none of these top 50 strains are 7 year survivors (2017 – 2023). 7 year survivor strains and strains with the most tests all have lower THC content (generally 20% - 25%) than that for the strains with the highest average THC % (generally over 30%). As a caveat, most of the highest THC % strains entered in 2022 or 2023, so they might not have had time enough yet to become more popular.

Figure 14