You are on page 1of 11

Dimacale 1

Maddee Dimacale
Econometrics, Dr. Dressler
December 11, 2018

Topic: How does pollination affect the productivity, defined as berry abundance, of

huckleberries?

Brief Summary

I am using my own data for this project from a summer research internship centered

around bumblebees and huckleberries. I am using my data on huckleberry phenology, netting,

and hand-pollination to be able to estimate huckleberry abundance using a simple linear analysis.

This data from my study focused on how bee pollination affects huckleberry productivity,

defined as berry abundance and wet weight, on the Flathead Reservation in Western Montana.

Half of the sample plants were covered in nets to prevent bee pollination and then half of both

the netted and unnetted plants were hand-pollinated. Therefore, there were 30 netted and hand-

pollinated plants, 30 unnetted and hand-pollinated plants, 30 netted and not hand-pollinated

plants, and 30 control plants with no treatment.

Bee populations have fluctuated in the last decade due to varying reason such as colony

collapse, which affects specific plants that rely upon bee pollination. Huckleberries are a

culturally significant plant to Native Americans on the Flathead Reservation and they have

impacts on ecology and economics. Huckleberries play a role in the economy since they are

edible and have other beneficial characteristics. The plant produces black/dark purple berries that

people consume that are similar to blueberries. They are included in a number of commercial

products such as pies, jams, drinks, and syrups. Areas such as Montana use these products as

tourist novelties because these fruits do not grow across the whole country. Montana’s
Dimacale 2

huckleberry industry had a revenue of $1.5 million in 1998 (Tom Howard 2016 Oct 2). The

question looked into was how a disturbance in bee populations in western Montana would affect

the production/productivity of huckleberries. This study also looks into if it be possible for

people to match the productivity of huckleberries with hand-pollination instead of natural animal

pollination. The goal of this project is to help people to understand the importance of pollinators

and how they can affect huckleberry productivity.

Similar Study

Asare, Eric, et al. "Economic Risk of Bee Pollination in Maine Wild Blueberry, Vaccinium

Angustifolium." Journal of economic entomology 110.5 (2017): 1980-

92. ProQuest. Web. 1 Oct. 2018.

A similar study was done on blueberries. Using OLS regression, Asare (2017) estimated

a model for wild blueberry yield using fruit set, a set of farming system dummy variables

representing high-, medium-, and low-input systems, native bee density, honey bee density, and a

production season dummy. This study was a good reference when conducting my project, but it

is not really comparable to my data. Huckleberries are not commercially farmed like blueberries

and other edible plants, so many of the studies I found would be difficult to compare

huckleberries to. Huckleberries are vastly under researched, making it more interesting to me to

focus my project on them.

The Empirical Analysis, Theory

Independent Variables

 “Net”
Dimacale 3

o Whether or not the huckleberry plant was covered by a net (blocks natural

pollination from bumblebees)

o Dummy variable with 0 being covered by a net and 1 being uncovered

o I think that there will be a negative relationship between being covered by a net

and berry abundance because without a net, bumblebees can pollinate the flowers.

 “Handpollinated”

o Whether or not the plant was hand-pollinated

o Dummy variable with 1 being hand-pollinated and 0 being not hand-pollinated

o I think there will be a positive relationship between hand-pollination and berry

abundance because if pollinators do not go to the flowers, then the flowers will

still be pollinated.

 “Phenology”

o Number of flowers and buds initially counted from first phenology data from

March

o I think that there will be a positive relationship because the flowers and buds

should continue through their phenology to become berries barring other factors.

Dependent Variable

 “Berries”

o Number of berries collected by the end of the growing season in July.

o I think that the berry abundance will be the most highly impacted by whether or

not the plant was covered by a net because pollinators have been found to be

extremely important to these plants.


Dimacale 4

Data Source

All data comes from independent summer research by Maddee Dimacale, Janene Lichtenberg,

and Dr. Antony Berthelote in 2018 on the Flathead Reservation in Montana supported by the

National Science Foundation, Salish Kootenai College, and the University of Minnesota. Data

collection was also helped by Dennis, Lichtenberg, Kendra Melanson, Sarah Zurkee, Rachel

Loder, and Morgan Schultz.

Classical Model Assumptions

Initial view of data to get intuition behind model

These initial graphs give me an idea that there may be a violation of an assumption with

my phenology variable (Figure 3).

Figure 1. Relationship between netting and berry abundance.


Dimacale 5

Figure 2. Relationship between hand-pollination and berry abundance.

Figure 3. Relationship between phenology data and berry abundance.


Dimacale 6

Assumption I: linear, correctly specified, additive error term

Berriesi = 0 + 2Net + 3Handpollinated + 4Phenology + i

Assumption II: Error term has a zero population mean

Ordinary Least Squares forces this assumption to be true.

Assumption III: Independent variables are uncorrelated with the error term

When comparing my error term to my independent variables, I did not see a pattern that

would indicate that they are correlated.

Figure 4. Relationship between error term and netting


Dimacale 7

Figure 5. Relationship between error term and hand-pollination

Figure 6. Relationship between error term and phenology


Dimacale 8

Assumption IV: No serial correlation of the error term:

My data is not time-series, it is cross-sectional so this does not apply to my model.

Assumption V: No heteroskedasticity

When running a studentized Breusch-Pagan test, I found that my model did have a

problem with heteroskedasticity as shown by the p-value of 0.00000007329. This means

that the error term must not have a constant variance throughout the range of each

independent variable.

studentized Breusch-Pagan test

data: REG

BP = 32.858, df = 2, p-value = 7.329e-08

Based on these results, there’s a 0.000007329% chance of being wrong if I reject the null

hypothesis, which is a type I error. This shows there is heteroskedasticity meaning that

the variability of one value is unequal across the range of values of a second variable that

predicts it. I need to use robust standard errors to correct this.

To fix it I did a t-test of coefficients, which gave me my new robust standard errors:

t test of coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -0.879374 0.948823 -0.9268 0.35595
Net 1.715655 0.961109 1.7851 0.07686
Handpollinated 1.127245 1.074423 1.0492 0.29628
Phenology 0.977998 0.089429 10.9361 < 2e-16 ***

Assumption VI: No perfect multicollinearity


Dimacale 9

To check for multicollinearity, I used variance inflation factors (VIFs) to see if any of my

variables are redundant. All of my VIFs were less than 5, showing that my model does

not violate assumption VI.

Netted/Unnetted Hand-Pollinated Phenology


1.143839 1.037566 1.181405

Assumption VII: Error term is normally distributed

This is not required by the OLS regression model.

Final Results

Summary results (including new robust standard errors):

Estimate Std. Error t value Pr(>|t|)

(Intercept) -0.87937 0.948823 -0.835 0.406

Net 1.71566 0.961109 1.340 0.183

Handpollinated 1.12724 1.074423 0.925 0.357

Phenology 0.97800 0.089429 24.382 <2e-16 ***

Residual standard error: 6.555 on 116 degrees of freedom

Multiple R-squared: 0.8646, Adjusted R-squared: 0.8611

F-statistic: 246.9 on 3 and 116 DF, p-value: < 2.2e-16

“Net”

All else equal, plants that were not netted would have 1.72 more berries than plants that

were netted. This is understandable because this plant flourishes when it is pollinated by

bumblebees.
Dimacale 10

“Handpollinated”

All else equal, plants that were hand-pollinated would have 1.13 more berries than plants

that were not hand-pollinated. This is understandable because the hand-pollination

ensures that the plant is successfully pollinated, which helps them grow.

“Phenology”

All else equal, plants will have 0.978 more berries for every one increase in number of

flowers or buds counted during phenology. This was the most statistically significant

variable as indicated by the low p-value. This is understandable because the flowers and

buds should continue to grow into berries in theory, but other factors may influence

whether it makes it to a berry (temperature, rain, etc.) that would lower the coefficient to

slightly less than 1.

Final Equation

Berriesi = -0.879 + 1.716(Net) + 1.127(Handpollinated) + 0.978(Phenology) + i

Goodness of Fit

According to the R2 value, my model’s independent variables can explain 86.46% of huckleberry

berry abundance (R2 = 0.8646). This means that some variables that affect berry abundance may

not be included, but there is a high goodness of fit.


Dimacale 11

Sources

Asare, Eric, et al. "Economic Risk of Bee Pollination in Maine Wild Blueberry, Vaccinium

Angustifolium." Journal of economic entomology 110.5 (2017): 1980-

92. ProQuest. Web. 1 Oct. 2018.

Tom Howard. (2016, October 2). From the Editor: Montana’s huckleberry industry is tiny, but it

packs a big punch. Retrieved from https://billingsgazette.com/business/columns/from-

the-editor/from-the-editor-montana-s-huckleberry-industry-is-tiny-but/article_937c787d-

a204-5801-9486-754a437c978e.html

You might also like