5 | Main Results and Findings from Sociologic Surveys
5.2 Hypotheses from classical qualitative sociological sur- sur-veys
A classical sociological qualitative and quantitative survey was realised in Autumn semester 2011-2012 and a new qualitative survey in Autumn 2012 that will be followed by a quan-titative questionnaire in Spring semester 2013. Below are few examples of hypotheses coming from classical qualitative sociological surveys:
Hypotheses from classical qualitative sociological surveys 159
• People display more propensity to develop a responsible behaviour toward energy savings if they have been educated to this behaviour from their childhood.
• If the government introduces a quota for electricity and quantities consumed above that quota would be billed more, people who approve this measure would modify their energy consumer behaviour.
• Eco-responsible people are less sensible to energy cost and would not change their behaviour to save money compared to people who so far did not behave as energy savers.
In order to prepare a usable input for energy models, these hypotheses have been quantitatively processed as follows:
• Determining the proportion of people who behave rationally even without the in-formation campaign
• Determining the part of people who are not behaving rationally but would change their behaviour if better informed
• Find the part of people who do not want to change at all
• Identify drivers and barriers to behavioural change in order to address them and remove them
• Find proxy parameters characterising the behaviour
A description of Social MARKAL is in Section 4.2 on page 123. For Social TIMES, in Section 4.3 on page 132. A comprehensive cookbook how to come from the results of a sociological survey to technical coefficients for Social MARKAL is in the Annex 8 on page 201.
However, besides these main results, we have obtained many secondary results in forms of hypotheses and descriptive statistics.
For example, in the first survey conducted between September 2009 and March 2010 on a sample of 393 respondents in the Geneva area (in cantons of Geneva, Vaud and in France), the questionnaire on the lighting bulbs consisted of 18 multiple choice questions, 1 open question and 8 socio-economic (ethnologic) questions. The survey is reproduced in Annex on page 177.
Typically, there are four types of questions, see Hoevenagel (1994):
1. Attitude - level of agreement to a statement measured using a Likert scale1 . 2. Behaviour - response to a particular situation or stimulus, measured precisely (for
example question: how many bulbs do you have at home? response: exact answer, ratio, interval).
3. Information - level of awareness regarding some facts, measured precisely (yes, no, do not know).
4. Hypothetical scenarios - circumstance, condition, or situation that might arise (for example question: if you were better informed, would you... possible answers: yes, no)
In case of testing bivariate hypotheses, we typically search to
• explore differences between classes on two variables (for example, differences in attitude and behaviour depending on age, education, income, etc.)
• explore relationships between two variables (for example, a relationship between an attitude such as in favour of sustainable development, and behaviour such as purchasing low-consumption bulbs)
An example of a hypothesis to explore differences between classes of two variables.
H0: Gender has no influence on the number of bulbs replaced by low-consumption bulbs during the last two years.
H1: Gender has an influence on the number of bulbs replaced by low-consumption bulbs during the last two years.
Definition of gender variable (nominal): Q20 Gender (man, woman)
Definition of bulb replacement variable "NumberOfReplacedBulbs" (ordinal on a trans-formed scale): Q8 During the last two years, how many incandescence bulbs did you replace by low-consumption bulbs? (exact number, none,1/4,1/2,3/4, all of them, do not know)
Reformulated hypothesis H0: The distribution of variable N umberOf ReplacedBulbs is the same across categories of variableGender.
1Likert scale, named after Rensis Likert, is a psychometric scale used to scaling responses in survey research, expressing the level of agreement or disagreement on a symmetric agree-disagree scale and capturing the intensity of feelings of the respondent for the studied item. Since it ican be defined as symmetric and equidistant, it may be treated either as ordinal or interval variable. Nunnally and Bernstein (1994) suggest to treat the variable as continuous if it contains at least 11 categories. In our surveys, we typically used an even number of categories to avoid neutral positions and force the respondent to take a position evaluated either as like or dislike for the studied item.
Hypotheses from classical qualitative sociological surveys 161 Test: Mann-Whitney U test.
Asymptotic significance: 0.061
Decision: Retain the H0 hypothesis at significance level 5%. Alpha risk 6.1% (type I error).
Thus it cannot be concluded that gender has a significant influence on behaviour of replacement of incandescent bulbs by low-consumption bulbs.
An example of a hypothesis to explore differences between classes of two variables.
H0: There is no relationship between turning off lights systematically after leaving a room and parental education to turn off lights.
H1: There is a relationship between turning off lights systematically after leaving a room and parental education to turn off lights.
Definition of "TurningLightsOff" variable (nominal): Q3 Do you systematically turn off lights when leaving a room? (yes, no)
Definition of "ParentalEducation" variable (nominal): Q5 Did your parents educate you to turn off lights when leaving a room? (yes, no, do not know)
Test: χ2 test.
Result: Asymptotic significance 0.000 (p < 0.00001)
Decision: Reject the H0 hypothesis at significance level 5% (in fact, at any significance level above 0.001%). Alpha risk ≈0.0%.
Thus it can be concluded that there is a statistically significant relationship between parental education to turn off lights when leaving a room and the behaviour to turn off lights when leaving a room.
χ2 computation is in tables 5.2, 5.3 and 5.5 on the next page.
Variable: TurningLightsOff
Observed N Expected N Residual Resid2 Resid2/Expect
No 113 196.5 -83.5 6972.25 35.482
Yes 280 196.5 83.5 6972.25 35.482
Total 393 70.964
Variable: ParentalEducation
Observed N Expected N Residual Resid2 Resid2/Expect
No 43 130.0 -87.0 7569 58.223
Yes 321 130.0 191.0 36481 280.623
Do not know 26 130.0 -104.0 10816 83.200
Total 390 422.046
Table 5.2: Frequencies: Q3 and Q5
TurningLigtsOff ParentalEducation
Chi-square 70.964 422.046
df 1 2
Asymp. Sig. 0.000 0.000
Table 5.3: Test statistics: Q3 and Q5. Continuing on the next page.
An example of a hypothesis to explore the relationship between two variables.
H0: The number of bulbs people have at home is not positively rank-correlated to the desired frequency of information.
H1: The number of bulbs people have at home is positively rank-correlated to the desired frequency of information.
Definition of "NumberOfInstalledBulbs" variable (ordinal): Q6 How many bulbs do you have at home? (exact number, 0-10, 11-20, 21-30, 31-40, 40+, don’t know)
Definition of "InformationFrequency" variable (ordinal): Q17 What frequency of infor-mation would you find useful for you? (once a year, 2-3/year, once a month, continuously in order not to forget, not at all, don’t know)
Tests: Kendall’sτb and Spearman’sρ Significance: 4.7 and 4.9%, respectively
Decision: Reject the H0 hypothesis at significance level 5%. Alpha risk 4.7 and 4.9%, respectively.
Thus, there is a significant relationship between the number of bulbs people have at home and their desire to be frequently informed about low-consumption bulbs, see Table 5.6.
Hypotheses from classical qualitative sociological surveys 163
observed TurningLightsOff ParentalEducation yes no total
yes 250 71 321
no 16 30 46
don’t know 14 12 26
total 280 113 393
expected TurningLightsOff ParentalEducation yes no
yes 228.702 92.298
no 32.774 13.226
don’t know 18.524 7.476
χ2 statistics TurningLightsOff ParentalEducation yes no
yes 1.983 4.914
no 8.585 21.272
don’t know 1.105 2.738
40.597 Table 5.4: χ2 statistics for questions Q3 / Q5
Probability p
0.05 0.01 0.001 0.0001 df 1 3.841 6.635 10.828 15.14
2 5.991 9.210 13.816 18.42
Table 5.5: Values ofχ2 for common p values and degrees of freedom 1 and 2
NumberOfInstalledBulbs InformationFrequency
Kendall’s NumberOfInstalledBulbs Correlation Coefficient 1.000 0.095*
tau_b Sig. (2-tailed) 0.047
N 295 292
InformationFrequency Correlation Coefficient 0.095* 1.000
Sig. (2-tailed) 0.047
N 292 390
Spearman’s NumberOfInstalledBulbs Correlation Coefficient 1.000 0.115*
rho Sig. (2-tailed) 0.049
N 295 292
InformationFrequency Correlation Coefficient 0.115* 1.000
Sig. (2-tailed) 0.049
N 292 390
Table 5.6: Test statistics: Q6/Q17.
* Correlation is significant at the 0.05 level (2-tailed).