As part of the real accountability data analysis for result project, also known as RADAR Project, we develop a set of tools including a sample size calculator for hazards server. The objective of this lesson is to introduce RADAR sample size calculator and how to use it. Sample size calculation is one of the most complex component people typically struggle during this step of a sever design and planning. To ease the calculation of sample size, there are existing online sample size calculator. However, most of them are not appropriate for household server as they do not account for a complex server design nor the convention from individual to household, and therefore may underestimate a sample size. In that respect, we develop an Excel-base and a web-base application to properly calculate the sample size for household server. The focus of this presentation will be on the web-base application available at the web address you are able to see on your screen. The RADAR sample size calculator is a free online platform. We usually use a step-by-step approach for people not familiar, we have a sample size calculation to correctly plan and calculate the sample size. The tool allow you to calculate the sample size for a given precision or a power-based options. For precision-based options, you will be able to calculate both of the required number of target populations and the required number of household to achieve the desired precision for your sample. Using the power-based option, you will be able to calculate both of the required number of target population and the required number of household based on the minimum detectable difference between two population group or two time point that the sample will be able to detect. On the other end, you can also calculate the expected level of precision of a desired number of household, or to calculate the minimum detectable difference or minimum detectable magnitude of change that a given sample allows to detect. Let's go through the sample size calculator and see how to use it. Starting first by the power-based option. For each option, you just have to choose the option and then click on "Next" to have access to next page. This is how the page looks like when you start using the sample size calculator. You have a tag on About that should provide you a few information related to the tool. A tag for the language that you are able to select either English or French for your exercise. Then there is also a tag where you are able to provide some feedback or comments regarding the tool. On the left side, you have the menu, the menu where you should fill out all the different parameters, input information needed to calculate the sample size. You also have a list of our indicators. There is a predefined list of indicators. We have a target population for each of them. The different indicators are grouped by type, according to continuum of care, starting from family planning, antenatal care, delivery, postnatal care, child preventive care, child curative health care, nutrition, nutrition impact, vector control, water and sanitation, poverty, child protection, sexual behavior, sexual health, gender, woman empowerment et cetera. Also you are able to customize this list, adding some indicators of interest which are not in this list. Let's try to see an example on how we can use an indicator so as to calculate a sample size. Let's pick first the contraceptive prevalence rate. This is our example. Then we can also add a second indicator also. But let's think, now there is an indicator that is not in this list but you want to customize the list. You just need to come and click on ''Manage Custom Indicators''. Then from this, now you are able to add an indicator. For instance, let's say that we are going to call this custom indicators example of custom indicators. This is what we are going to use as our custom indicator, and we just need to enter and then come and apply here to now this list. Here you are able to see that we have both indicators that we selected for this contraceptive prevalence rate, and custom indicators. For each of them, we needed to provide some input data in order to be able to calculate the sample size, or clicking on next to have access on where to customize and provide with different information. We should provide input information for each of the indicators starting first by contraceptive prevalence rate. Right now, we have a red sign here. Let's say that we're interested for this indicator to target women age of 15-49 here. I have to select a lower age bound 19-49, and the upper bound is 49. I'm okay with that. Then I have to get some input information for the parameter from previous server implemented in that area or similar area. This is a helper menu where you can also try to figure out on how to complete each section. For this one, I have to provide the number of women age 19-49 here in union which is my target population. This number I'm able to get it from a previous server. Let's say that I got the number of 890 for the number of women. Then those women have been interviewed from, let's say, a 956 households. We have a household response rate of 0.96. This is a response rate for these indicators. Once I fill out the different input of the parameters, the sign become now green saying that I'm okay. We have these first indicators, they have now to move to the second one. This is the custom indicator, also. For the custom indicator, also the tool allows you to select the target population. If you know your target population, you can also select from here. But if the target population is not suggested by the tool, you can also enter your own target population selecting this options where you are able now to come here and then change the name of your target population, saying that, okay, maybe a target population, custom indicators, for instance. So that can be, for instance, the target population. But in case the population is already in the default list of target population from the tool, you can select from here. Let's say that it's in this tool and it's this second option. We have known to provide information relating to the age and also, for this target population, we also need to provide information related to a reference period. Let's say that maybe this is an indicator relating to adolescent girls between 15-19 years old. Then reference period will be adolescent girls with a live birth in the last two years. I'm going to provide this information as well so that now I have my target population that is properly customized for this indicators as well. But I didn't have to prevent none of the different input value. Here, we may have, from previous studies, let's say, 250 adolescent which had been interviewed from the same number of household, that is 956. Then we will be seeing more. Response rate, that is 0.96. Then I'm going to view this indicators as well. Once I'm done, I have now a greener sign here. I can move now to the next step letting the two parameters or letting to the estimated frequency and so. As we said in previous sessions, sample size calculation, you are able to calculate the number of household or to also calculate based on some parameters or to calculate the minimum detectable difference based on a given number of household. Let's see first this example where we want to calculate the required number of household given our difference parameters. Here for prevalence contraceptive rate, we should provide the rate for the estimate for the reference group. That can be a baseline or a reference group. Also for the second arm that can be the end line. Let's say here we have the base line and end line. We baseline we have a frequency of let's say 80 percent and you have now to end up say frequency of 25 percent for modern contraceptive rate. From previous study, we found that of this indicators are exact effect of 2.1. Then for the second indicators, we were able to find that the estimator frequency at the baseline is 65 percent. But we want to improve this indicators and reach a proportion of 75 percent so with design effect of two. That mean that for the first indicators we have a magnitude of change of about seven percent and for the second one we have a magnitude change of about 10 percent. Let's now see how will be the result in term of required number of individual and also the required number of household. That is our last step. Here now this is how my results looks like. It means that if I want to be able to detect a difference or a change over time of about a seven percent minimum difference of change I want to be able to detect. Based on both proportion and also design effect I should interviewed about 1,134 women age 19-49 year in union. To be able to get this number of women I should aim to to interview 1,269 household in total. For my second indicators also to be able to detect a minimum magnitude of change of 10 percentage point between my baseline and my end line. We've design effect of two I should also interviewed about 659 adolescent girl age of 15-19 with a live birth in the last two years. But to be able to get this number of adolescent I should interviewed about 2,623 householder. This is not my final outcome in term of sample size for my study. I can print out or save this table of all the result for any over-proposal, but since as I said, we're also able to see what is the minimum detectable difference based on a given sample size. So this is now what is sample we're going to do. We have now to select a this option of difference. Then from this menu, we should also provide our different number of household and see what can be the minimum detectable differences. Let's say we [inaudible] on 2,000 householder and we want to see what can be the minimum detectable differences based on 2,000 household and also keeping the same parameters we already entered. If you go down in result, we are able to see that we were 2,000 household. We will be able to have a minimum detectable difference of six percentage point between the baseline and end-line for this study, or between Group 1 and Group 2. This is the required number of woman I need for that. Also for the same number of household, I will be able to detect a minimum difference of 11 percentage point between the two groups or between the two time point. This is what we are able to do for the Power-base example. Let's see now how it's look like also for the Precision-base example. We have to go to the main menu, choose a mode, the mode now will be our Precision-base options. I have also to enter the different information. But let's keep the information I already enter to go faster. But if you want to enter new information, you just need to clear here and you will have now to enter new indicators and also new input. But let's keep both indicators for this example as well. Also let's keep everything we've already enter as assign target population here as well and move to next step. In the next step now, let's now try to see what are the number of household we are able to get for different percentage estimator. For this example we are going to use the most conservative assumption of estimated frequency, that is 50 percent for both indicators. Then for desired precision, let's say that we want to use a five percentage points as desired precision. For design effect let's use two as design effect for both indicators and see what will be the sample size based on those different assumption. In that case, it means that, to have some estimator with a desired precision of five percentage point of a contraceptive prevalence rate, we have to interview 860 households for this first indicator. For the second indicator also, we have to interview up to 3000 household to be able to have a desired precision of a five percentage point. You are able to see that for different indicators and also different target populations, we end up with different sample size. In case that our study, we don't have enough resources to go and interview up to 3000 household, we cannot [inaudible] figure out what can be the different change we can make to our different parameters to see. In case that we come back and say that, okay, we want to see for the adolescents, instead of the 5 percentage point, you want to have precision of 10 percentage pointer and see what will be now of the sample size. In that case, were able to see that, we should now him to interview about 765 household to have a desired precision of 10 percentage point. As we previously point out in the previous sessions, you're able to see that, depending on the indicators and also depending of the target population, we are able to end up with different sample size. We have also to see that based on the available resources we have for our study, what can be the good trade off, in terms of precision or in terms of magnitude of change or difference between group we want to detect. We are also able to see what is the precision base on a given number of households, which is something quite similar to what we previously just show for the Power-base example. So this is some of the potential for this tool that you can use for different proposal in term of sample size calculations. Thank you for your attention.
