ISSUE: Impact measurement

FACTS: There are 2 communities: (i) tourism oriented community that exploits more of forest resources and (ii) non-tourism oriented that exploits less of forest resources. The target of measurement is: "what is the level of forest exploitation by each community?"

DATA SOUGHT: There are three types of data: (i) quantitative, (ii) ordinal, and (iii) nominal. If you use written survey to collect data, try to collect all three types of data so that you analysis would involve multiple tests, thus, the result (final report) would be data rich and methodologically interesting. If use survey questionnaires, avoid this type of Likert scale: 1, 2,3,4,5. Try to use scale that includes zero: (0, 1, 2, 3) where 0 = none, 1 = low, 2 = medium, and 3 = high. As for intensity level measurement, leave that to statistical test on parameters.

DESIGN: Let Y = dependent variable (effect or effect level of forest exploitation). Let Xi = independent variables, i.e. x1, x2, .., xn. You may run simple regression between Y and X, thus:

(1)   Y = B0 + B1X1

The test statistic is a simple t-test given by:

(2)   tr = r[sqrt(n-2) / sqrt(1 - r2)]

... where r = Sx / Sy.

If using multiple independent variable, multiple regressions is given by:

(3)   Y = B0 + B1X1 + B2B2 + ...+BmXm

The test statistic to verify statistical significance is:

(4)   F = MSR / MSE

(4.1)   MSR = SSR / p     where p = number of Xi

(4.2)   MSE = SSE / n - p - 1

(4.3)   MST = SST / n -1

(4.4)   SST = Σ(Yi - Y*)     where Y* = mean of Yi

(4.5)   SSE = Σ(Yi - Y^)     where Y^ = Y-hat

(4.6)   SSR = Σ(Yi^ -Y*)

LATENT EFFECT MEASUREMENT: With the multiple regression, try to confirm whether there are "interaction effect" among the independent variables: X1, X2, ... by employing the latent effect model thus:

(5) Y = B0 + B1X1 + B2X2 + B3X1X2 + e

... where B0 = intercept; and X1X2 the interaction effect between X1 and X2 to produce Y---which is generally not measurable in conventional multiple regression equation. The parameter for the latent effect is determined by:

(6)   B3 = B^ -B1 - B2 - B0

(7)   B^ = Y* - (B2 - B0) - B1

The test statistic to verify the significance of the interaction is given by:

(8)   TLatent = (B1 - B2) / sqrt(C / D)

(8.1)   C = n1SE12 + n2SE22

(8.2)   D = n1 + n2 - 2

The designed thus described is for structural modeling using continuous data. You may also want to measure the effect level of forest exploitation in a dichotomous form: discrete data type. That would involve different statistical test.

EXPLOITATION DEFINED AND MEASURED: If the answer sought for exploitation of forest issue is Yes = exploit and No = not exploit, then the response may be coded as Yes = 1 and n = 2. The probability of exploitation follows:

(9) p = (s + 1) / n + 2

... where s = number of Yes = 1. The probability of no exploitation is simply q = 1 - p. The test statistic, assuming that the data tends towards normal distribution, is given by:

(10)   Z = [(x / n) - p] / sqrt(pq / n)

... where x = target number of observation to be tested, i.e. exploitation counts where the probability of x is given as:

(11)   P(X) = ABC

(11.1)   A = (n! / (n - X)!X!

(11.2)   B = pX

(11.3)   C = qn-X

IMPACT MEASUREMENT: In order to measure impact, you need to allow the time to run from a starting period to a specified period subsequent to the starting time, then measure the significance of the changes. There are two ways of testing the impact where the data is dichotomous: (i) Two counts Poisson and (ii) Z-test for correlation proportion.

(i) Two Counts Poisson Test: Assume there there are two time period of the measurement t1 and t2. The target measurement is forest exploitation. Let the quantity or frequency or the count of exploitation be N and for the two time periods then the counts are N1 and N2. The average frequencies are: R1 = N1 / t1 and R2 = N2 / t2. The test statistic is given by

(12)   Z = (R1 - R2) / sqrt((R1 / t1) + (R2 / t2)

The null hypothesis is R1 = R2, meaning there is no impact or no different between the two communities.

(ii) Z-Test for Correlation Proportion: A second test is based on the chi-square distribution. It involves polling the answers among the targeted respondents in two time periods. Construct 2 X 2 table. Let the column be first poll and the row be the second poll for the answer Yes & no exploitation of forest.

                    YEScolumn          NOcolumn

YESrow      a                             b

NOrow       c                              d

Total counts: N = (c + c) + (b+d) + (a + b) + (c + d). The test statistic is given by:

(13)   Z = (b - c) / Nσ

(13.1)   σ = sqrt[((b+c) - (b - c)2 / N) / N(N - 1)]

I hope this could give you some helpful pointers to get your project started. Cheers.

REFERENCES: See link below.

http://fcc-statistics.wikispaces.com/file/view/100+Statistical+Tests.pdf

ISSUE: Impact measurement

FACTS: There are 2 communities: (i) tourism oriented community that exploits more of forest resources and (ii) non-tourism oriented that exploits less of forest resources. The target of measurement is: "what is the level of forest exploitation by each community?"

DATA SOUGHT: There are three types of data: (i) quantitative, (ii) ordinal, and (iii) nominal. If you use written survey to collect data, try to collect all three types of data so that you analysis would involve multiple tests, thus, the result (final report) would be data rich and methodologically interesting. If use survey questionnaires, avoid this type of Likert scale: 1, 2,3,4,5. Try to use scale that includes zero: (0, 1, 2, 3) where 0 = none, 1 = low, 2 = medium, and 3 = high. As for intensity level measurement, leave that to statistical test on parameters.

DESIGN: Let Y = dependent variable (effect or effect level of forest exploitation). Let Xi = independent variables, i.e. x1, x2, .., xn. You may run simple regression between Y and X, thus:

(1)   Y = B0 + B1X1

The test statistic is a simple t-test given by:

(2)   tr = r[sqrt(n-2) / sqrt(1 - r2)]

... where r = Sx / Sy.

If using multiple independent variable, multiple regressions is given by:

(3)   Y = B0 + B1X1 + B2B2 + ...+BmXm

The test statistic to verify statistical significance is:

(4)   F = MSR / MSE

(4.1)   MSR = SSR / p     where p = number of Xi

(4.2)   MSE = SSE / n - p - 1

(4.3)   MST = SST / n -1

(4.4)   SST = Σ(Yi - Y*)     where Y* = mean of Yi

(4.5)   SSE = Σ(Yi - Y^)     where Y^ = Y-hat

(4.6)   SSR = Σ(Yi^ -Y*)

LATENT EFFECT MEASUREMENT: With the multiple regression, try to confirm whether there are "interaction effect" among the independent variables: X1, X2, ... by employing the latent effect model thus:

(5) Y = B0 + B1X1 + B2X2 + B3X1X2 + e

... where B0 = intercept; and X1X2 the interaction effect between X1 and X2 to produce Y---which is generally not measurable in conventional multiple regression equation. The parameter for the latent effect is determined by:

(6)   B3 = B^ -B1 - B2 - B0

(7)   B^ = Y* - (B2 - B0) - B1

The test statistic to verify the significance of the interaction is given by:

(8)   TLatent = (B1 - B2) / sqrt(C / D)

(8.1)   C = n1SE12 + n2SE22

(8.2)   D = n1 + n2 - 2

The designed thus described is for structural modeling using continuous data. You may also want to measure the effect level of forest exploitation in a dichotomous form: discrete data type. That would involve different statistical test.

EXPLOITATION DEFINED AND MEASURED: If the answer sought for exploitation of forest issue is Yes = exploit and No = not exploit, then the response may be coded as Yes = 1 and n = 2. The probability of exploitation follows:

(9) p = (s + 1) / n + 2

... where s = number of Yes = 1. The probability of no exploitation is simply q = 1 - p. The test statistic, assuming that the data tends towards normal distribution, is given by:

(10)   Z = [(x / n) - p] / sqrt(pq / n)

... where x = target number of observation to be tested, i.e. exploitation counts where the probability of x is given as:

(11)   P(X) = ABC

(11.1)   A = (n! / (n - X)!X!

(11.2)   B = pX

(11.3)   C = qn-X

IMPACT MEASUREMENT: In order to measure impact, you need to allow the time to run from a starting period to a specified period subsequent to the starting time, then measure the significance of the changes. There are two ways of testing the impact where the data is dichotomous: (i) Two counts Poisson and (ii) Z-test for correlation proportion.

(i) Two Counts Poisson Test: Assume there there are two time period of the measurement t1 and t2. The target measurement is forest exploitation. Let the quantity or frequency or the count of exploitation be N and for the two time periods then the counts are N1 and N2. The average frequencies are: R1 = N1 / t1 and R2 = N2 / t2. The test statistic is given by

(12)   Z = (R1 - R2) / sqrt((R1 / t1) + (R2 / t2)

The null hypothesis is R1 = R2, meaning there is no impact or no different between the two communities.

(ii) Z-Test for Correlation Proportion: A second test is based on the chi-square distribution. It involves polling the answers among the targeted respondents in two time periods. Construct 2 X 2 table. Let the column be first poll and the row be the second poll for the answer Yes & no exploitation of forest.

                    YEScolumn          NOcolumn

YESrow      a                             b

NOrow       c                              d

Total counts: N = (c + c) + (b+d) + (a + b) + (c + d). The test statistic is given by:

(13)   Z = (b - c) / Nσ

(13.1)   σ = sqrt[((b+c) - (b - c)2 / N) / N(N - 1)]

I hope this could give you some helpful pointers to get your project started. Cheers.

REFERENCES: See link below.

http://fcc-statistics.wikispaces.com/file/view/100+Statistical+Tests.pdf

Aon,

Important will be the definition of "exploit", from a tourism perspective. Your questionnaire would need to include categories of exploitation, ie use of wood for firewood, impact on forest due to transport or traveling  (transport physical damage, hiker impact), forest impact as result of pollution, construction or building impact, biodiversity influences, etc. correlating these with tourist and non-tourist activities. 

FC

Just a consideration to keep in mind:  It is difficult to think of purely forest dependent and tourism dependent communities in a geographic sense.  Even in places where many benefit from tourism, there may be individuals or groups within the community  who (a) are/have been forest dependent (b) get little or no benefit from tourism.  Look, for example, at Vancouver Island in Canada, where both tourism and forest industry are important, but to different people within the same general area. 

As highlighted, the first requirement is to clearly define the term "exploitation".  Are you referring to dependence on forests for livelihoods here?  It is important to see if the forests serve as a natural asset for tourism in the study area.  Here, the recreational value of the forest is taken into account.  I have attached a study I undertook in regard to  forest-based ecotourism in relation to Sri Lanka.  Hope you will be able to get some insights!

Sometimes getting people to draw maps (e.g. of where trees have been cut by outsiders, where firewood and nontimber forest products are gathered, where land has been cleared for agriculture) is more effective than "word" or "scale" questions.  The maps can in many cases be converted into quantitative data and may be more accurate than asking people to estimate numbers abstractly.

You may want to read the following articles to assist you:

https://workspace.imperial.ac.uk/environmentalpolicy/public/Beauchamp09.ExecutiveSummary.pdf

http://www.forestsmonitor.org/uploads/2e90368e95c9fb4f82d3d562fea6ed8d/Note_Executive_Forum4_Eng_for_web_SL_edits_final_1.pdf

And please read the attached file.

Thanks to all

Aon, I think that you got a lot of ideas, both quantitative (from Paul) and methodological (like drawing maps, by Robert). But you were asking about tricky question.

How can the behavior of local community differ w.r.t. forest exploitation depending whether it is from "tourist" or "non-tourist" area?

I guess that forest is mostly exploited for logging (used in local construction) and firewood (for cooking food on grill, including one for tourists). The presence of tourists create higher demand of locals for forest exploitation by small business.

I think that the questionnaire should be oriented on this small business, owners of restaurants and hotels. It may include the questions:

a) What fuel are you using for cooking (with firewood as one option in answer)

b) Are you using local wood for construction works in hotels and on beaches?