Professional Documents
Culture Documents
INFERENTIAL STATISTICS
Concerned with analysis of data from a sample leading to
predictions or associations (interferences) about the target
population
Page 1 of 10
2 types of making inferences about the population Reliability coefficients
1. Hypothesis testing – making decisions as to whether or - From probability distribution tables (Z, t, X2)
not a statement about the population is true based on - Based on desired confidence level
the evidence from the sample
2. Estimation – using a statistic to approximate the Reliability coefficient
Confidence level
parameter (Z value)**
90% 1.64
When we do hypothesis testing and estimation to infer about 95% 1.96
the target population, we are also considering the random 99% 2.58
error. ** for proportions
Standard Deviation = √
Page 2 of 10
Interpretation: Confidence Interval for Risk Ratio
15% (point estimate) of the elderly in Northern India have Risk Factor (+) disease (-) disease Total
poor or fair quality of life. The authors are 95% (confidence (+) A B A+B
level) confident that the true percentage of adults who have (-) C D C+D
fair or poor quality of life could be between 12% and 18%
(confidence interval – upper limit and lower limit).
Relative Risk =
Where:
Confidence Interval for Odds Ratio
P=
Risk Factor (+) disease (-) disease Total
(+) A B A+B
Difference between two
independent means √ (-) C D C+D
( ̅ 1- ̅ 2 )
Odds Ratio =
Where:
sp =√
exp ( √[ ] [ ] [ ] [ ]
Solution:
Relative Risk = = = 0.581
sp =√
The result 0.581 means that it is preventive.
Less likely to have MI if person is taking aspirin
sp =√ = 17.7
exp ( √
SD error of difference
= √ exp ( √
Page 3 of 10
Estimation STEP 3: DETERMINATION OF APPROPRIATE TEST STATISTIC
The width of the confidence interval indicates the
imprecision of the sample estimate. Supposed we find a 10% difference in cure rate between two
The wider the interval, the less precision treatments
The width of the confidence interval depends on: This observed difference could be:
Sample size - A true difference
Variability of characteristics being studied - Due to chance: in reality there’s no difference
The degree of confidence required - Due to bias: defects in the study design
High degree of confidence = Wider confidence interval Is the difference in the two groups due to chance?
Must rule out “chance” as explanation of the difference
by applying significance testing
HYPOTHESIS TESTING
A significance/statistical test estimates the likelihood
Steps that an observed study result (ex. difference between 2
1. State the null and alternative hypothesis. groups) is due to chance
2. Set the level of significance. “An appropriate chosen test of significance allows us to
3. Determine the most appropriate test statistic. calculate the probability of obtaining a difference
4. Apply the test to the data. Determine either the critical between sample estimates as large as the observe
value or the p-value. difference, given that the Ho is true.”
5. Make a statistical decision.
6. State the conclusion. Factors affecting choice of test statistic:
Research question
STEP 1: STATE THE HO AND HA Scale of measurement
Independent or paired measurement
Hypothesis – statement about the population
Two types Paired/related groups
Null hypothesis (Ho) – no difference, no association, no Matched groups by age, gender or any characteristics
relationship; hypothesis to be tested One group tested twice (pretest, post-test)
Alternative hypothesis (Ha) – a statement that disagrees Crossover trial – one group will receive both treatments
with the H0 Example: 50 patients will receive Treatment A during the first
week then the same patients will receive Treatment B on the
One tailed – the direction is either stated as being better or worse second week.
Two tailed – the direction is not stated
Independent groups – two groups are not related
Example: Example: Clinical trial – 100 were randomized to either
Study objective: To determine association of oral Treatment A or Treatment B (50 patients will be given
contraceptive use with increase in serum cholesterol level. Treatment A, 50 patients will be given Treatment B)
HO: µOC = µnon-oc
Mean cholesterol level of OC users is equal to mean Measurement Scale
cholesterol level of non-OC users Goal Interval/
Nominal Ordinal
Ratio
HA: µOC µnon-oc
Compare one
Mean cholesterol level of OC users is not equal to mean group to a Kolmogorov- One sample T-
Binomial Test
cholesterol level of non-OC users. hypothetical Smirnov Runs test test
HA: µOC µnon-oc value
Compare two
Mean cholesterol level of OC users is greater than mean a. Sign test
related or McNemar Paired T-test
cholesterol level of non-OC users. b. Wilcoxon test
paired groups
Compare two a. Fisher’s
a. Median test
independent Exact (2x2) T-test
b. Mann-Whitney
STEP 2: SET THE LEVEL OF SIGNIFICANCE groups b. Chi-square
Compare more
than two Two-way
Level of Significance – the probability of rejecting a true null Cochran’s Q Friedman
related or Anova
hypothesis paired groups
Compare more
Truth in the population than 2 One-way
Statistical Decision Chi-square Kruskal-Wallis
independent Anova
Ho is true Ho is false
groups
Reject Ho Type 1 error (α) Power (1-β) Quantify
Do not reject Ho Type II error (β) Spearman’s rank Pearson’s
association
Contingency correlation correlation
between two
coefficient coefficient
α – usual values: 0.05, 0.01, 0.10 variables
Probability of committing a Type 1 error
Chance of rejecting a true Ho Tests for interval/ratio data are parametric tests while tests
Level of significance for nominal/ordinal data are non-parametric tests.
Page 4 of 10
Parametric tests Example #2:
One that makes assumptions about the parameters (defining Twenty-four experimental animals with Vitamin D deficiency
properties) of the population distributions from which one’s were divided equally into two groups. Group 1 received
data are drawn Vitamin D enriched diet while Group 2 received no treatment.
Assumptions of Parametric Tests At the end of the experimental period, serum calcium
a. Normal distribution (mg/100 mL) levels were measured. Is there treatment
b. Interval or ratio scale of measurement effect?
c. Random selection
d. Equal variances (or known ratio of variances) in Study Design: Experimental Study
comparison groups
Objective: To determine if serum calcium Vitamin D enriched
If one of the assumptions is seriously violated, diet has an effect on calcium levels.
nonparametric tests are used instead.
Compare:
Nonparametric tests How many groups? 2 (Group 1 Vit. D & Group 2 Placebo)
No assumptions about shape of distribution Are the groups independent or related? Independent
Appropriate test: What are we comparing? Calcium levels (mg/100 mL)
a. When the data represent an ordinal or nominal scale Measurement scale? Ratio
b. When a parametric assumption has been greatly violated
c. When the nature of the distribution is unknown Measurement Scale
Goal Interval/
Nominal Ordinal
Example: Ratio
A study was done to determine if inadequate monitoring is Compare one
Kolmogorov-
group to a One sample
associated with death among hospitalized asthma patients. Binomial Test Smirnov Runs
hypothetical T-test
test
value
35 asthma patients who died were individually matched for Compare two a. Sign test
Paired T-
sex and age with 35 control patients who had been related or McNemar b. Wilcoxon
test
discharged alive from the same hospital. The inadequacy of paired groups test
monitoring of all patients while in the hospital was assessed. Compare two a. Fisher’s a. Median test
independent Exact (2x2) b. Mann- T-test
groups b. Chi-square Whitney
Study Design: Case-Control Compare more
than two Two-way
Cochran’s Q Friedman
Objective: To determine association of monitoring death related or Anova
among hospitalized asthma patients. paired groups
Compare more
than 2 One-way
Compare: Chi-square Kruskal-Wallis
independent Anova
How many groups? 2 (Cases, Controls) groups
Are the groups independent or related? Matched groups Quantify Spearman’s
Pearson’s
What are we comparing? Exposure status: Monitoring association rank
Contingency correlation
Measurement scale? Nominal (Inadequate, Adequate) between two correlation
coefficient
variables coefficient
The test value based on the sample will be determined if 1. State the null and alternative hypothesis.
it lies in the acceptance area or rejection area. HO: There is no association between taking aspirin and MI
The test value has a corresponding p-value. HA: There is an association between taking aspirin and MI
P- value – the probability that the observed result could have 2. Set the level of significance.
arisen by chance, probability of obtaining a result as extreme α = 0.05
as or more extreme than the one observed if the Ho is true. It
is compared to α or level of significance. 3. Determine the most appropriate test statistic.
∑
Usual value of α = 0.05
Page 6 of 10
Statistical significance vs. Clinical Significance 24. What is the most appropriate statistical test to determine if the
null hypothesis in number 23 can be rejected?
A. Chi-square test
B. T-test
C. Paired T-test
D. McNemar test
#AllFor2019 28. The following statement is true for level of significance except:
A. Probability of committing a Type 1 error
Transcription Team 2019 B. Probability of rejecting a true null hypothesis
Transcribed by: Trisha Mae M. Bongcales C. Probability of not rejecting a false null hypothesis
References: Lecture, PPT, Internet D. None of the above.
Remarks: Thank you, Danica Bautista,
for a copy of the recordings 29. If the confidence interval of the prevalence ratio includes the
of this lecture! value of one, then one can conclude that:
A. There is significant association between outcome and independent
Less time to mourn variable.
B. There is an increased risk of developing the outcome if exposed to
More time to grind
the independent variable.
#LABAN2019 C. There is a lower risk of developing the outcome if not exposed to
the independent variable.
D. There is no evidence from the data to conclude an association
Past E questions: between the independent and the outcome variable.
Batch 2012
21. Researchers observed that bone mineral density (BMD) is related 30. Not rejecting a null hypothesis means:
with tea drinking among older women. The researchers used __ to be A. The null hypothesis is true.
able to conclude that older women who drank tea had higher BMD B. The null hypothesis is false.
measurements (g/cm2) than those who did not drink tea. C. There is enough evidence to conclude the alternative hypothesis.
A. Chi-square test D. There is insufficient evidence to conclude the alternative
B. T-test hypothesis.
C. Paired T-test
D. McNemar Test 31. Parametric tests assumptions include the following except:
A. Positive or negative skewness of the distribution
22. The researchers can use the ___ to test the null hypothesis that B. Equal variances between groups
proportion of hormone replacement therapy users among tea drinkers C. Interval or ratio scale of measurements
is equal to the proportion of hormone replacement therapy users D. Normal distribution
among non-tea drinkers.
A. Chi-square test For nos. 32-37.
B. T-test A survey was conducted to determine the prevalence of UTI among
C. Paired T-test pregnant mothers in a small community. These pregnant mothers
D. McNemar test were also monitored and birth weight of their babies was recorded.
23. A group of students wanted to determine if there is a relationship The researchers wanted to know if birth weight of babies of mother
between depression and internet addiction. They plan to compute the with UTI would be different from babies of mothers without UTI
prevalence ratio to quantify the association. What would be the most during pregnancy.
appropriate null hypothesis for their study?
A. Mean level of depression among internet addicts is not equal to 32. What is the most appropriate null hypothesis?
the mean level of depression among non-internet addicts. A. Pregnant mothers with UTI will have babies whose birth weights
B. Mean level of depression among internet addicts is equal to the will be lower than those whose mothers did not have UTI.
mean level of depression among non-internet addicts. B. Pregnant mothers without UTI will have babies whose birth
C. The proportion with depression among internet addicts is not equal weights will be lower than those whose mothers had UTI.
to the proportion with depression among non-internet addicts. C. There is no difference in mean birth weight of babies between
D. The proportion with depression among internet addicts is mothers who were UTI positive and UTI negative.
equal to the proportion with depression among non-internet D. There is a difference in mean birth weight of babies between
addicts. mothers who were UTI positive and UTI negative.
Page 7 of 10
33. What could be the most appropriate alternative hypothesis? D. It is 1.43 times less likely that those who leave school without
A. Pregnant mothers with UTI will have babies whose birth weights qualifications watched more television than those who leave school
will be lower than those whose mothers did not have UTI. with qualifications.
B. Pregnant mothers without UTI will have babies whose birth
weights will be lower than those whose mothers had UTI. Risk ratio
C. There is no difference in mean birth weight of babies between Relative risk
mothers who were UTI positive and UTI negative. Risk of developing a disease (outcome) in a group of people who
D. There is a difference in mean birth weight of babies between were exposed to a risk factor relative to a group who were not
mothers who were UTI positive and UTI negative. exposed to the risk factor
34. What is the most appropriate test to determine if mean birth Choices B. and D. are wrong because of the terms “less likely”
weight of babies of mothers with UTI is not equal to the mean birth which are not descriptive of risk ratio. Choice C. is more correct than
weight of babies of mothers without UTI? choice A. because the former is more specific to what the study was
A. ANOVA looking for.
B. Chi-square
C. Paired T-test 32. Given the 95% CI of the risk ratio 1.43, the authors’ statistical
D. Independent’s T-test decision was:
A. Reject the null hypothesis that there is no association between
35. What is the computed p-value of the statistical test for testing the watching TV and poor educational achievement.
null hypothesis in question #32? B. Do not reject null hypothesis that there is an association between
A. 0.7497 watching TV and poor educational achievement.
B. 0.4668 C. Do not reject the null hypothesis that there is no association
C. 0.3226 between watching TV and poor educational achievement.
D. 0.9784 D. Reject the null hypothesis that there is an association between
watching TV and poor educational achievement.
36. At 0.10 level of significance the null hypothesis is
A. Rejected RR > 1 = there is an increased risk of developing disease (outcome)
B. Not rejected if one is exposed to the factor
C. Accepted
D. Discarded Choices B. and C. are wrong because the null hypothesis should be
rejected because there is an association. Choice D. is also wrong
37. The author’s conclusion should be as follows: because the null hypothesis should state that there is NO association
A. Infant’s birth weight is associated with the mother’s having UTI
during pregnancy. 33. Given the 95% CI of the risk ratio 1.43, the authors can conclude
B. Results did not provide evidence that mean birth weight of babies that:
whose mothers were UTI positive is different than that of babies A. Watching TV is a preventive factor for poor educational
whose mothers were UTI negative. achievement.
C. Mean birth weight of babies whose mothers were UTI positive is B. Study did not provide evidence of association between watching
equal to the mean birth weight of babies whose mothers were UTI TV and poor educational achievement.
negative. C. Watching TV is not associated with poor educational achievement.
D. Results did not provide evidence that mean birth weight of D. Watching TV is a risk factor for poor educational
babies whose mothers were UTI positive is equal than that of achievement.
babies whose mothers were UTI negative.
RR > 1 = suggests exposure may be associated (“cause”) with the
Batch 2018 disease or outcome
31. Hancox et al. conducted a study to determine the association of
childhood viewing during childhood with poor educational 34. This can be used to determine if there is significant association
achievement (ARCH PEDIATR ADOLESC MED/VOL 159, JULY between gender and poor educational achievement:
2005 pp 614-618). The study abstract is as follows: A. T- test
B. Chi- square test
Background: Excessive television viewing in childhood has been C. ANOVA
associated with adverse effects on health and behavior. A common D. McNemar’s test
concern is that watching too much television may also have a Measurement Scale
negative impact on education. However, no long-term studies have Goal Interval/
Nominal Ordinal
measured childhood viewing and educational achievement. Ratio
Objective: To explore these associations in a birth cohort followed Compare one
up to adulthood. group to a Kolmogorov-
Binomial Test One sample T-test
Design: Prospective birth cohort study hypothetical Smirnov Runs test
value
Setting: Dunedin, New Zealand.
Compare two
Participants: Approximately 1000 unselected individuals born c. Sign test
related or paired McNemar Paired T-test
between April 1, 1972, and March 31, 1973. Ninety-six percent of the d. Wilcoxon test
groups
living cohort participated at 26 years of age. Compare two c. Fisher’s
Main Outcome Measures: Educational achievement by 26 years of age. e. Median test
independent Exact (2x2) T-test
f. Mann-Whitney
Results: The mean time spent watching television during childhood groups d. Chi-square
and adolescence was significantly associated with leaving school Compare more
without qualifications and negatively associated with attaining a than two related Cochran’s Q Friedman Two-way Anova
university degree. Risk ratios for each hour of television viewing per or paired groups
weeknight, adjusted for IQ and sex, were 1.43 (95% confidence Compare more
than 2
interval [CI], 1.24-1.65) and 0.75 (95% CI, 0.67-0.85), respectively Chi-square Kruskal-Wallis One-way Anova
independent
(both, P=0.001). The findings were similar in men and women and groups
persisted after further adjustment for socioeconomic status and early Quantify
childhood behavioral problems. Spearman’s rank Pearson’s
association
Contingency correlation correlation
between two
coefficient coefficient
The reported risk ratio of 1.43 (95% CI 1.24-1.65) would mean that: variables
A. It is 1.43 times more likely that those who leave school without
qualifications watched more television than those who leave school The variable “gender” was measured using a scale dichotomized to
with qualifications. male/female meaning it is utilizing the nominal scale and the
B. It is 1.43 times less likely that a person would leave school variables being compared is independent of each other (male and
without qualifications for each hour increase in television viewing. female). Referring to the table above, the correct answer is Fisher’s
C. It is 1.43 times more likely that a person would leave school exact or Chi-square.
without qualifications for each hour increase in television
viewing.
Page 8 of 10
35. This can be used to determine if there is significant difference in 41. If the confidence interval of the Prevalence Ratio includes the
mean duration of television viewing between those with university value of one, then one can conclude that there is:
degree and those without university degree: A An increased risk of developing the outcome if exposed to the
A. Chi- square test independent variable.
B. McNemar’s test B A lower risk of developing the outcome if not exposed to the
C. T- test independent C Significant association between outcome and
D. ANOVA independent variable.
D. No evidence from the data to conclude an association between
Mean duration of television viewing utilizes the ratio scale and the the independent and the outcome variable.
variables being compared are independent of each other (those with
and without university degree). Referring to the table on #34, the 1 is equal to “no association” or the null value
correct answer is T-test.
42. Mutura et al conducted a case-control study to determine factors
36. The authors also classified educational achievement into four associated with defaulting from treatment among tuberculosis
categories: no qualifications, school certificate, post-school patients. Table 1 shows the gender and treatment observer of the
qualifications and university degree. This would be the most study population.
appropriate test if the authors want to determine significant
differences in mean duration of television viewing among the four Table 1 Gender and Treatment Observer of study population (Source:
classifications: secondary data, n = 1978)
A. McNemar’s test Characteristic Treatment Completed Total OR P-
B. ANOVA defaulters treatment % (95% CI) value
C. Chi- square test (n = 945) (n=1033)
D. T- test N% N%
Gender
Mean duration of television viewing utilizes the ratio scale and there Male 561 (59.4) 547 (53.0) 1108 1.3 0.01
(56.0) (1.08-
are more than two independent groups (no qualifications, school Female 384 (40.6) 486 (47.0) 870 1.55)
certificate, post-school qualifications, and university degree). (44.0)
Referring to the table on #34, the correct answer is One-way ANOVA Treatment Observer
Health care 266 (28.1) 248 (24.0) 514 1.24 0.04
37. The width of the confidence interval is affected by the following, worker (26.0) (1.014-
EXCEPT: Household 679 (71.9) 785 (76.0) 1.464 1.52)
member (74.0)
A. Confidence level
B. Sample size
One of the study’s null hypothesis states that:
C. Variance of the variable
A. There is no relationship between gender and defaulting
D. Size of the target population
treatment among tuberculosis patients.
B. There is a relationship between gender and defaulting treatment
Width of the confidence interval
among tuberculosis patients.
Sample size
C. There is an association between completing and defaulting
Variability of characteristics being studied
treatment among tuberculosis patients.
The degree of confidence required
D. There is no association between completing and defaulting
treatment among tuberculosis patients.
38. The p-value of a statistical test is the probability of observing:
A. The difference given the null hypothesis is true
Null Hypothesis (Ho)
B. The difference given the alternative hypothesis is true
State of no difference, no association, no relationship
C. No difference given the alternative hypothesis is true
Hypothesis to be tested
D. No difference given the null hypothesis is true
Deals with target population
P-value
The probability that the observed result could have arisen by chance
Choices B. and C. are wrong because they show a relationship and
The probability of obtaining a result as extreme as or more extreme than association respectively. Choice D. is also wrong because the study is
the one observed if the Ho is true determining the factors associated with defaulting and NOT
completing treatment
39. Not rejecting a null hypothesis means:
A. The null hypothesis is false 43. This is the appropriate test to determine if the null hypothesis can
B. There is enough evidence to conclude the alternative hypothesis be rejected:
C. The null hypothesis is true A. Chi- square
D. There is no enough evidence to conclude the alternative B. ANOVA
hypothesis C. T- test
D. Mann- Whitney
Condition Significance Decision Conclusion
p-value <α Test is significant Reject Ho Conclude HA Both variables used are in the nominal scale. Since the variables
There is no being compared are independent of each other (defaulted treatment
Test is not Do not reject sufficient and completed treatment). Referring to the table on #34, the correct
p-value >α
significant Ho evidence to answer is Fisher’s exact or Chi-square.
conclude HA
44. The following is true based on what is presented in Table 1:
40. Assumptions of parametric tests include the following, EXCEPT: A. Both variables “gender" and “treatment observer” are
A. Normal distribution significantly associated with defaulting treatment.
B. Positive or negative skewness of the distribution B. Both variables ―gender" and ―treatment observer‖ are not
C. Equal variances between groups significantly associated with defaulting treatment.
D. Interval or ratio scale of measurements C. "Gender" is the only variable significantly associated with
defaulting treatment.
Assumptions of parametric tests D "Treatment observer" is the only variable significantly associated
Normal distribution with defaulting treatment.
Interval or ratio scale of measurement
Random selection
Condition Significance Decision
Equal variances (or known ratio of variances) in comparison groups
p-value <α test value is in
Test is significant Reject Ho
the rejection area
p-value >α test value is in
Test is not significant Do not reject Ho
the acceptance area
Gender: 0.01 < 0.05 = test is significant
Treatment observer: 0.04 < 0.05 = test is significant
Page 9 of 10
45. The null hypothesis ―There is no association between gender and 56. Considering the data presented in Table 3, the result of the
defaulting treatment‖ is: hypothesis testing for the null hypothesis ―Gender is not associated
A. Rejected with depressive symptoms‖ is— The test is:
B. None of the choices A. Significant, p- value is greater than 0.05
C. Not rejected B. Significant, p- value is less than 0.05
D. Accepted C. Not significant, p- value is greater than 0.05
D. Not significant, p- value is less than 0.05
Refer to table on #44.
0.01 < 0.05 = reject Ho Refer to table on #44.
46. The following statement is true: 57. Considering data presented in Table 3, the authors can conclude that:
A. Male and female tuberculosis patients have the same treatment A. Boys and girls have the equal risks of developing depressive symptoms.
default rate. B. Boys are more likely to have depressive symptoms compared to girls.
B. The odds that the defaulters are female are greater than them to be C. Gender is not associated with having depressive symptoms.
D. Girls are more likely to have depressive symptoms compared to boys.
males.
C. The odds that the defaulters are male are greater than them to
Just infer from Table 3 (Table on #55).
be females.
D. Gender is not associated with defaulting treatment.
58. The variable ―depressive symptoms‖ was measured using a scale
and dichotomized to with/without by using a cut- off score of 17. The
The percentage of male defaulters are 59.4% whereas female
appropriate statistical tool to test the hypotheses ―Parental education
defaulters are 40.6% so choice B. is the correct answer.
is not associated with depressive symptoms‖ is:
A. Chi- square
47. The null hypothesis ―There is no association between treatment
B. Mann- Whitney
observer and defaulting treatment‖ is:
C. T- test
A. Accepted
D. ANOVA
B. None of the choices
C. Not rejected
Different variable used but same ratio with #55
D. Rejected
59. Considering data presented in Table 3, the result of the hypothesis
Refer to table on #44.
testing for the null hypothesis ―Parental education is not associated
0.04 < 0.05 = reject Ho
with depressive symptoms‖ – The test is:
A. Significant, p-value is greater than 0.05
48. The following statement can be part of the conclusion:
B. Not significant, p-value is less than 0.05
A. Treatment observer is significantly associated with defaulting
C. Significant, p-value is less than 0.05
treatment.
D. Not significant, p- value is greater than 0.05
B. Completion of treatment is less likely among patients with health
care workers as the treatment observer compared to those observed
Refer to table on #44.
by household member.
C. Completion of treatment is more likely among patients with health
60. Considering data presented in Table 3, the authors can conclude
care workers as the treatment observer compared to those observed
that:
by household member.
A. Those whose parents are of high education are more likely to have
D. Treatment observer is not significantly associated with
depressive symptoms than those whose parents are of low education.
defaulting treatment.
B. The risks of having depressive symptoms are the same for all
categories of parental education.
Refer to table on # 39.
C. Those whose parents are of low education are more likely to have
depressive symptoms than those whose parents are of high education.
55. Wirback et al conducted a longitudinal study to determine the
D. The odds ratios for the association of parental education with
association of childhood social factors and risk of depressive
depressive symptoms are not statistically significant.
symptoms. Table 3 below shows some of the study results:
Refer to table on #39.
Table 3. Odds ratios of depressive symptoms in adolescence,
measured as DSM- IV criteria based by social factors, n=1880
Page 10 of 10