Introducing precisely
an R package for estimating sample size based on precision rather than power
1 / 39
2 / 39
precisely lets you estimate sample size based on precision instead of significance
3 / 39
precisely lets you estimate sample size based on precision instead of significance
calculate n for common measures, like risk difference and risk ratio
4 / 39
precisely lets you estimate sample size based on precision instead of significance
calculate n for common measures, like risk difference and risk ratio
Work in R or on the web
5 / 39
6 / 39
7 / 39
What is precision?
8 / 39
What is precision?
Differences: width of the CI
9 / 39
What is precision?
Differences: width of the CI
Ratios: upper bound/lower bound
10 / 39
You need:
desired precision
11 / 39
You need:
desired precision
proportions exposed/unexposed
12 / 39
You need:
desired precision
proportions exposed/unexposed
group ratio
13 / 39
You need:
desired precision
proportions exposed/unexposed
group ratio
coverage
14 / 39
the precisely R package
🎯 Calculate sample size for precision
15 / 39
the precisely R package
🎯 Calculate sample size for precision
install from GitHub:
remotes::install_github("malcolmbarrett/precisely")
15 / 39
Estimating n for risk difference
library(precisely)n_risk_difference()16 / 39
Estimating n for risk difference
library(precisely)n_risk_difference( precision = .08,)17 / 39
Estimating n for risk difference
library(precisely)n_risk_difference( precision = .08, exposed = .4, unexposed = .3,)18 / 39
Estimating n for risk difference
library(precisely)n_risk_difference( precision = .08, exposed = .4, unexposed = .3, group_ratio = 3,)19 / 39
Estimating n for risk difference
library(precisely)n_risk_difference( precision = .08, exposed = .4, unexposed = .3, group_ratio = 3, ci = .90)20 / 39
Estimating n for risk difference
library(precisely)n_risk_difference( precision = .08, exposed = .4, unexposed = .3, group_ratio = 3, ci = .90 )## # A tibble: 1 x 8## n_exposed n_unexposed n_total risk_difference## <dbl> <dbl> <dbl> <dbl>## 1 524. 1573. 2097. 0.1## # … with 4 more variables: precision <dbl>,## # exposed <dbl>, unexposed <dbl>,## # group_ratio <dbl>21 / 39
Estimating n for odds ratios
precisely::n_odds_ratio( precision = 2, exposed_cases = .4, exposed_controls = .3, group_ratio = 4)22 / 39
Estimating n for odds ratios
precisely::n_odds_ratio( precision = 2, exposed_cases = .4, exposed_controls = .3, group_ratio = 4)## # A tibble: 1 x 9## n_cases n_controls n_total odds_ratio precision## <dbl> <dbl> <dbl> <dbl> <dbl>## 1 171. 685. 857. 1.56 2## # … with 4 more variables: exposed_cases <dbl>,## # exposed_controls <dbl>, group_ratio <dbl>,## # ci <dbl>23 / 39
| measure | differences | ratios |
|---|---|---|
| risks | ✔️ | ✔️ |
| rates | ✔️ | ✔️ |
| odds | ✖️ | ✔️ |
24 / 39
Estimating precision given sample size
precision_odds_ratio( n_cases = 500, exposed_cases = .6, exposed_controls = .4, group_ratio = 2)25 / 39
Estimating precision given sample size
precision_odds_ratio( n_cases = 500, exposed_cases = .6, exposed_controls = .4, group_ratio = 2)## # A tibble: 1 x 9## precision odds_ratio n_cases n_controls n_total## <dbl> <dbl> <dbl> <dbl> <dbl>## 1 1.55 2.25 500 1000 1500## # … with 4 more variables: exposed_cases <dbl>,## # exposed_controls <dbl>, group_ratio <dbl>,## # ci <dbl>26 / 39
Vignettes and more: precisely.malco.io
27 / 39
What's the baggage?
29 / 39
What's the baggage?
precision == probability of estimate being true?
30 / 39
31 / 39
What's the baggage?
precision == probability of estimate being true?
The ghost of statistical testing shall always haunt us
32 / 39
33 / 39
34 / 39
precisely lets you estimate sample size based on precision instead of significance
35 / 39
precisely lets you estimate sample size based on precision instead of significance
calculate n for common measures, like risk difference and risk ratio
36 / 39
precisely lets you estimate sample size based on precision instead of significance
calculate n for common measures, like risk difference and risk ratio
Work in R or on the web
37 / 39
Appendix
38 / 39
39 / 39