Propensity Score Matching - Science topic

..

The tow models (PSM and endogenous switching regression) are the impact analysis models which can be used to analyze the impacts of an intervention. So, in what conditions each model can be applied?

Hello everyone! I'm seeking a comprehensive understanding of how to handle confounding variables when comparing two groups based on the presence of a specific variable. Should I use propensity score matching or multivariable logistic regression for this purpose?

I am conducting research in which I want to investigate the effect of tax incentives on research and development intensity in a group of firms. I have access to the data of a survey that:

I'm having a little trouble choosing the right econometric method for this research.

#microeconometrics

Hi,

I am working on analysis of a study where I need PSM matching . Kindly suggest a free software for PSM analysis. TIA.

Dear all,

I am seeking guidance regarding the appropriate method for analyzing a binary outcome in an observational clinical research study. The primary outcome is a binary variable. One main independent variable needs to be tested for its effect on the outcome, while other variables serve as theoretical confounders.

I am considering several options: traditional logistic regression, propensity score matching using nearest-neighbor (potentially sacrificing sample size), or matching with inverse-probability treatment weights (retaining the entire sample size). Your valuable suggestions would be highly appreciated, and any alternative methods are also welcome.

Thank you sincerely for your input.

Best regards,

Suppadech Tunruttanakul

Propensity score matching (PSM) and Endogenous Switching Regression (ESR) by full information maximum likelihood (FIML) are most commonly applied models in impact evaluation when there are no baseline data. Sometimes, it happens that the results from these two methods are different. In such cases, which one should be trusted the most because both models have their own drawbacks?

I am working on propensity score matching, honestly i didn't understand the intuitive mechanism of kernel algorithm as a method of matching based on the propensity score. In other words, how the conterfactual outcome is obtained using this method.

I have seen papers where PSM has been performed using cross sectional and panel data. I want to know if PSM can be used for time series data too.

I also have a question that which quantitative method should one use for analysing the impact of a policy intervention. The dataset is time series in nature.

Dear colleagues, I am conducting a quasi experimental analysis focused on the Italian labor market. More specifically, I would like to evaluate the impact of a specific reform over the probability (after some specific time period, e.g., 1 year) of being employed (or not) after the implementation of the reform. Therefore, the outcome variabile is a dichotomous variable equal to 1 if the individual is employed after the specific time period, 0 if not. In the literature, as far as I know, quasi-experimental studies are conducted using mainly continuous response variables. My idea is to rely on a Propensity Score Matching (PSM) approach. However, I would like to ask you for some suggestions and works to consult in order to properly undertake this analysis of mine. Thank you in advance.

After having run the propensity score analysis in R, I need to conduct a sensitivity analysis in the same software. Kindly suggest which R-package should I use.

The outcome variable in case of the propensity score matching analysis is very often a continuous variable. I am trying to conduct a propensity score matching analysis. My outcome variable for which I am trying to measure the average treatment effect on the treated is a five point likert scale. Can I still use the PSM method? If yes, is there any change in estimation techniques.

I am interested to compare two groups i.e, credit access users vs non-users.

What steps I must follow?

psmatch2 credit_access (sex_hh age_hh family_size distance collateral attitude farm_size educ_recode marital_recode), kernel outcome( production_in_cmv ) bwidth(0.25) common logit ate

In my study, I am using propensity score matching to balance the effects of covariates on the impact of prednisolone on death outcomes among COVID-19 patients. 92 covariates have been considered, including demographic factors, signs and symptoms, other drugs, laboratory tests, and vital indicators. A problem arises when I remove one of the variables (the ALT). This changes the final results significantly.

How can I ensure whether I should remove that variable?

I am comparing the effect treatment vs no treatment on health outcome in my retrospective longitudinal study. As subjects were probably not randomly allocated to the treatment group, I am using propensity score matching to asses treatment effect.

I have managed to get the ATE in Stata but I would like to build a table of descriptive variables (e.g age, height, weight, blood pressure etc) to compare the treated vs non-treated groups after PSM matching. Can anyone please help/advise how can I do this in Stata

I use Propensity score matching for two different kinds of treatment (dummy and continuous treatment), which I will use each treatment in separate regression. So I converted the continuous treatment (the board of director number) into the dummy, and used the Mahalanobis matching within propensity score calipers. However, I'm not sure how to support my argument, especially after reading the advantages and disadvantages of this method? (I use Rstudio) is this method is acceptable in business research?

Also, I'm not sure that other method methods like GPS will be the same effective as the Mahalanobis matching within propensity score calipers to achieve covariate balance for panel data?

I am conducting an impact study, which is a "pure natural experiment" type. However, the sample size of the treated group is somewhat higher that the comparison group. Does this ratio difference affect the estimates of propensity score matching model result? If so, what proportion is better?

Is it possible to use the Synthetic Control Method in the firm-level study?

Currently, I am trying to measure the impact of environmental regulation (state-level emission trading scheme) on a firm's green innovation. I have used DID and PSM-DID but want to use SCM with firm-level data, is it applicable? I found most of the paper used state-level data. Expecting your kind opinion in this regard.

It is intuitive to me that in a 1:1 proportion of treatment: control groups sample size would be problematic if a treatment observation fails to have a corresponding match from the control observations. I understand by increasing the number of control observations we reduce the probability of not finding a match to a treatment observation. Is it possible to have a treatment observation number that is more than the number of control observations?

can t-test be use as alternative analyses for propensity scores matching ?

I use PSM (propensity score matching) to analyze the treatment effect of my bond pricing data, and the result is interesting. Then I realized that the outcome variable is skewed which means that the average treatment effect may come from a few outliers. I went through similar literature which uses PSM on financial data but figured that it is not a required pre-processing step and previous literature did not do transformation of their outcome either. I am wondering if normal distribution of outcome variable is required for PSM.

Please kindly advise. Thank you in advance.

What is the most acceptable method to measure the impact of regulation/policy so far?

I only know the Difference-in-Difference (DID), Propensity Score Matching (PSM), Two-Step System GMM (for dynamic) are common methods. Expecting your opinion for 20 years long panel for firm-level data.

What`s the minimum sample of treated and non-treated observations for a study that uses a combination of Propensity Score Matching (PSM) and Difference in Difference (DID)?

I searched several articles online, and I could not find any "rule" or something that states what could be the minimum sample size of treated and non-treated observations on a study that uses PSM and DID approach as a combination. 

:

1) psmatch2 TreatmentVar (list of Xs), outcome neighbor(1) common

2) psmatch2 TreatmentVar (list of Xs), outcome common

then run the below command:

attnd outcome Treatment , pscore()

The second way of estimating ATT gives considerably higher t-stat.

Please suggest which is the right way to estimate ATT using psmatch2.

Note: I cannot use pscore command as my x list contains categorical variable

I'm planning to conduct an impact assessment using a cross-sectional data. So, from Endogenous Switching Regression Model (ESRM) and Propensity Score Matching (PSM), which is the most preferable model? Does Endogenous Switching Regression Model has merits over Propensity Score Matching?

I construct a propensity score matching 1:1 analysis, is it possible the result comes out to have better AUC (area under curve) after the matching (better AUC than the original data), is this correct? (Most of the study I've read, none of the AUC is better than the original data)

I'm currently exploring the opportunity of conducting an NMA of observational studies only. very rarely NMA is done using observational studies as the transitivity assumption is unlikely to be fulfilled because these studies will have different protocols, methods of patient selection and settings ..etc.

I'm aware of the methods that are used to incorporate non randomized studies into a network of RCTs and the existence of many approaches for that such as regression adjustment, propensity score matching techniques ..etc. but I'm not sure if those techniques can be implemented in an NMA of observational studies only.

So, do you think that an NMA is applicable in the context of observational studies only? and if so, how can we fulfil the transitivity assumption? can proper statistical tests address that?

I used to my research database the value 0.1 as cases difference as maximum, a lot of data were exclude.

While using Propensity Score Matching Approach for an impact assessment of a programme inyervention, what is the acceptable ratio of the sample size for paticipant and non-participant?. Some have advocated for ratio 1:3 for participant to non-participant to allow for easy matching. I need your input please. The sample size of program participant in my study is 2,300. Thanks

I am analysis a data using propensity score matching and path analysis. However, one of the assumptions in selecting matching variables for propensity score matching is that they should not affect selection into treatment and also not affected by the treatment variable (endogeneity problem and unconfoundedness assumption) (Caliendo & Kopeinig, 2008).

1. Am I making an analytical mistake (violating the assumption) if I run both propensity score matching and normal regression (OLS using outcome variable as dependent variable or probit regression using treatment variable as dependent) in the same study? Because I am using the same matching variables as independent variables in the OLS and probit.

2. Is it analytically correct to run a path analysis in addition to propensity score matching using the same matching variables? I am running the path analysis to trace the path towards the outcome variables.

I don't know when I should use 1:1 matching or 1: to many * .

Also re * how many is 'many'? how do I know how many I need?

Is this based on the number of similar variables having the same propensity scores? Therefore if 3 in the control group have similar propensity scores to 1 in treatment score then it would be 3:1

Not sure if I understood correctly, pls let me know

Hi!

Is there a way in either STATA (preferably) or SPSS to do a Propensity score matching on a dataset where the outcome variable is time dependent?

I am doing an observational study looking at whether exposure /non exposure to a certain medication prior to surgical treatment has an effect on the outcome. The dataset consists of roughly 12000 patients, follow up is from 8yrs to 6 months, the outcome variable is therefore time dependent.

I have currently performed cox proportional hazard analysis looking into exposure with several pertinent covariates.

When trying to perform a PS match in stata it seems STATA only perform a PS matching based on whether or not the outcome occurs, I have not found a way to include the time to event.

All help is appreciated!

I´d like to test whether a priori defined edges and communities change across two severity groups of inpatients with dissociation. We seperated groups via subthreshold and matched in ratio 1:1 for age and gender to the sample of high dissociators. But how to deal with the fact that less variance can be explained in patients with low dissociation. Can we adjust matching ratio to 1:2?

happy for any help,

Philipp Göbel

#networkanalysis #networkcomparison #symptomicsframework #severitygroups

I am conducting a retrospective study to evaluate two approaches to treatment. I need simplified guidance regarding the application of propensity matching; i.e How and which program to use.. I saw some youtube videos, however, I am still confused about how to start and what program to use.

Thank you in advance.

Dear All,

I have Panel Data fits difference in difference

I regress the (Bilateral Investment Treaties-BIT) on (Bilateral FDI). BIT: is dummy taking 1 if BIT exists and Zero Otherwise. While Bilateral FDI: Amount of FDI between the two economies. Objective: Examine If BIT enhances Bilateral FDI?

The issue is : - Each country have started its BIT with another pair country at a fixed time (different from the others): NO Fixed Time for the whole data.

Is it necessary to interpret the results of Probit while using Propensity Score Matching (PSM) for impact evaluation?

I am studying the impact of having a data breach on the tone/financial reporting quality of the 10ks and to avoid confounding variables, I investigated that propensity score matching was the solution. However, I don't know or dispose of the variables that affect the choice of treatment i.e being breached but I have the variables that "control" the outcome i.e Tone. Does anyone have proposition??

Hi,

I am using STATA to do propensity score matching on a large sample (n around 1000 or more).

Only way I know is the psmatch2 command, generate 1:1 matched pairs and identify them and run a univariable cox regression in those identical pairs. In the process I lose a lot of data.

I was wondering if anyone knows a way to use propensity scores as covriates or knows how to stratify the propensity score in blocks and use almost all data to run the regression.

Thank you in advance.

Dhaval

Hi All,

I Used Propensity Score matching, however I am confused at one point.

1- In first step I run the probit model

2- In second step I checked the treated and untreated by using pscore

3- I am confused How Can i run the regression again by using only matched sample?

I need this kind of result as I am attaching this figure.

My research topic is: The impact of remittances on household welfare in Kyrgyzstan. I have two panel datasets for the years 2013 and 2016, (the data on migration and remittance receiving is included). The welfare of families is measured through consumer and productive asset indices.

I face the problem of choosing an appropriate method for empirical analysis. According to my adviser, the best solution is propensity score matching (PSM). Moreover, he suggested to consider more than one year to see a dynamics in how much families are better off from receiving remittances or vice a versa.

However, according to what I have already read about the PSM, most researches apply it only to one year, not two or more. I am confused how to use them for the purpose of my research and this specific method of analysis.

Suggest me the best research papers you have gone through for understanding the propensity score matching (PSM) in social science/behavioural research.

How do I specify matching ratio in SPSS propensity score matching? I'm trying to match 3 controls to 1 patient. There doesn't seem to be that option.

I searched up a bit and found some extensions using R but I ran through the steps and it kept saying that the Integration R plug-in has not been installed. Is there any way round?

If it's not possible in SPSS, I'd appreciate suggestions for other softwares too. I have 0 programming experience / experience in R though.

Thanks

propensity score matching analysis is mostly done for observational data. Is that possible to do a for case control , retrospective data or not?

Hello everyone! I would like to develop a propensity score matching for three treatment groups using STATA. The psmatch2 function is apparently not suited for more than two treatment matches.

Any suggestions?

After propensity score matching if we are getting sample size of group A around 20 times of other group B. Baseline characteristics are properly balanced after PS matching. But what about the conclusions drawn on this kind of unbalanced data (in terms of sample size)?

Na = 680000 and Nb=36000. Is there any guidance or literature available for this?

I have become accustomed to matching groups with each other (a treatment group with a created control group) and I do this in R, but I would like to know if each individual case in the treatment could be acceptably paired with a case in the control.

For example, if you have a treatment group of those who took the drug, and you create a control group using a series of reasonable variables, then one could quite easily have two groups that are balanced. However, say we were also interested in the time (i.e. date) when the people in the treatment group took the drug. Could I then match, on a one to one basis, each case from the treatment with a case from the non-treatment and essentially give the non-treatment cases the same time (i.e. date) as their matched treatment counterpart?

Another way of asking this question is, would it be theoretically sound for me to move away from comparing groups to actually comparing individuals in the groups (matched VS non-matched)?

Thanks a lot.

Michael

I am trying to install the R-plugin for Propensity score matching for SPSS and each time I try the plugin doesn't work. I'm not doing something right, I guess.

I have been researching propensity score matching to create a control group to a few schools that have received an intervention (N=25). I have a lot of covariates I could match on but have not found any research about what would be a maximum given the size of the sample pool I could use. I figured a 1-to-1 matching technique would be useful to get a control group of 25 schools but I am not sure how many variablesI could use to get the closest match of 25 schools not receiving intervention. I realize theory is important; for example, including poverty index or school population. But I want to know how many variables could maximize the propensity score given the potential reservoir. Is there a formula that you can add a covariate for every 100 samples available in the reservoir? So, if you had 200 schools available as potential controls, you could only use 2 variables within your propensity score matching regression.

I have conducted a PSM using SPSS and generated new matched data with propensity scores. My original data set was based on a HH questionnaire with 367 respondents. I wanted to generate a control group out of this data. I had a "treatment" administered on 217 and the rest 150 did not have this treatment. I ran a rather successful PSM and got some good results or so I think.

1. Logit estimate returned a Pseudo- R2 of .147 and Log Likelihood of 484.016; Chi Square of 14.454 and Prob>Chi2 <.0001; I had seven variables and six of which significantly predicted the probability of being treated.

2. The mean of propensity scores were:

Treated: n= 217 Mean: .63743 Min: .13063 Max: .89690

Control: n=76 Mean: .57739 Min: .26269 Max: .82095

3. I determined a common support region as: .13063 - .82095. This restriction had serous consequence on my data. My final matched data for analysis came down drastically to just 152. (76 Treated vs 76 Control). Here is my question:

Is this a normal occurrence in PSM analysis? is it normal to have such a huge drop in the data points? What might be the problem or otherwise in the subsequent analysis (ATT analysis)?

Dear all,

I have a data set, where 80% of total respondents are coming under treated group and rest of them are in control group. In this situation, using PSM technique can be appropriate for estimating unbiased Average Treatment Effect (ATE)? and please suggest any other methods and research work regarding this to solve this problem.

Many thanks in advance

Can one use inverse probability weighted studies, weighted propensity score models and propensity score matched observational studies together in Metanalyses?

I have a large data set in which many variables have missing data. I have run multiple imputation to produce 20 imputations. I have split each data file out and have run propensity score matching in each of them in SPSS via Thoemme's plug-in. I have done the same thing in each data set and have 20 different outputs produced where I have a table describing the means treated, means control, SD control, and standard mean difference. I need to combine the results from these 20 imputations to show stat 1-20 and the standard deviations for each imputation before matching and then show stat 1-20, the propensity score, and stat 1-20 for each imputation after matching. I then need to combine these results into one to generate the parameters and standard errors. If anyone could tell me how to do this with the output that I have, I would greatly appreciate it.

I am doing research on colon cancer Want to compare two different treatments of survival in a retrospective data. I want to do propensity score matching to adjust for the bias. I am using inverse propensity score weight method. I calculated inverse propensity score weight from propensity score. I was able to get odds ratio by using this weighted variable in SPSS. But when i am trying to calculate KM curve I am getting the following error code "No statistics are computed because nonpositive or fractional case weights were found". Any suggestions of how to proceed? 

Experts in statistics recommend the value of propensity score matching for study participants in two arms that did not have balanced number and baseline characteristics. I am a bit confused on how to match for sample and characteristics. What technical steps can be followed for propensity score analysis?

For right comparison of two groups (cases and control), I underwent propensity score matching (PSM) accounting 8 variables which showed significant difference between the two groups: one of them is sex.

After PSM, cox proportional hazard model was also performed to evaluate association with survival.

In order to maximize the number available for analysis, PSM was done with control to case ratio=4 (K=4). However, in comparison of baseline characteristics, two groups showed significant difference in sex even after the matching (P=0.015).

I assume that this is due to high control to case ratio, and the two groups had shown dramatic difference in male proportion (45% vs 80%) before PSM.

Would this significant difference after matching be a potential point for major criticism?

I am planning to include sex for multivariate cox proportional hazard model, considering this significant difference.

I have a quasi-experimental design in which propensity score matching is a good idea because being in the treatment group was self-selected. Do I have to include all potential confounders into the propensity score or can I decide to have one (that is also theoretically related to the motivation to participate in the treatment, in this case: high energy consumption levels for participating in an energy feedback program) as a covariate in the model after having propensity-matched samples that are almost equal on the other confounders? And what would be the explanation/justification of doing so? This model makes more sense in my view but I'm lacking a methodological explanation for having some variables as covariate and some as a propensity score.

Thanks for any help!

Does the control variable must always be a non-user of irrigation for PSM to work?

Dear All,

I work with a STATA 14 on the estimation of the ATET. I have population data (N=900,000) and for the sake of transparency, I want to keep the large sample. I get estimates with psmatch2 for PSM and kelner matching (after a day), however I can never get estimates for NN (command teffects nn match). The strange thing is that if I make the sample smaller, the estimates work.

Basically, I have the following questions:

1) Is there any way how to use different command to estimate teffects nn match?

2) Do you have any idea, how to estimate this effect with teffects nn match (I have been working in this issue really a lot of days and I have no clue, how to overcome this).

Many thanks for your replies in advance.

What are the different parameters we should include in an analysis of PSM? Is it possible to apply the technique in a secondary data? What should be the ideal sample size for both treatment & control group then?

I have a data set where treatment group is larger than control group. Can I use propensity score matching?

I have temporal twitter data, and I want to calculate propensity score for the treatment and control group. The problem is, the treatment happened at different time for different user, and I want to compare aggregated values for covariates. For example, for user a1, treatment occurred after six months of account creation, for user a2, treatment occurred after eight months. I want to take the average number of tweets posted by those users before the treatment (for first six and eight months respectively), and find similar users in the control group who posted similar number of tweets for six and eight months. I have other covariates (e.g. average number of hashtags used). Is propensity score matching a suitable approach for this? If so how can I do that?

One approach came to my mind is to manually match one covariate at a time. This is similar to finding nearest neighbors, but for one dimension at a time. But not sure if this is a valid approach. If not what else I can do?

In our experiment, the treatment is the first tweet of an user that was retweeted more than a certain number of times. We want to compare for example, average number of tweets posted before and after the treatment by a user in treatment group with someone in the control group.

UPDATE:

Another way I think might be reasonable to find match using propensity score for one treated person at a time. First for a1 find someone in the control group based on propensity score who has similar average tweet (and other time dependent covariates) for first six months. Then for a2 find someone in the control group based on propensity score who has similar average tweet (and other time dependent covariates) for first eight months, and so on. Problem is in this approach each time there is only one data point from the controlled group, and often there is a perfect separation, so MLE does not have any solution. So instead of using MLE, may be I can use Nearest Neighbor approach to find closest person in control group in terms of the covariances, for each treated individual at a time.

I have generated a paired dataset in SPSS using propensity score matching. I now want to run a logistic regression with a different DV. Do I need to control for the propensity score? Or can I just run a logistic regression and use the treatment variable for the selection variable? Please help!!

Given that propensity scores are often derived via logistic regression, why bother? Why not just do a good old fashioned logistic regression analysis in the first place to adjust for confounding? It seems that propensity score matching can only be as good as the regression analysis that spawned it, so why not reduce the number of 'moving parts' and go for the simpler option of just using a regression analysis?

I'm carrying out a case control study

Hi folks, I am using r to run multiple imputation on propensity score matching, followed by my outcome regression model which is a HLM model. I read a thread: https://stats.stackexchange.com/questions/257789/manipulating-data-for-propensity-score-matching-following-multiple-imputation-wi that talks about the use of package mira to get the average regression coefficients from the imputed datasets but the package doesn't work in my occasion. Anyone knows how I could get the pooled regression estimate if the outcome regression is not a simple regression but a complicated one such as HLM? Thanks.

We intend to do research in East Africa to determine the impact of new tools to support farmers of a certain size, using for example mobile telephones or internet. we intend to use propensity score matching, a statistical matching technique (Wikipedia) that attempts to estimate the effect of an intervention by accounting for the covariates that predict receiving the treatment. PSM attempts to reduce the bias due to confounding variables that could be found in an estimate of the treatment effect obtained from simply comparing outcomes among units that received the treatment versus those that did not.

Can any one please send me any document or article shows a clear explanation for using the PS matching in spss  ( explanation of the detailed output of the ps matching)?

Thanks in advance 

Hi all!

I am to conduct an ex post impact evaluation for the first of a two-phased irrigation project. Given that, the pipeline approach was used so the comparison is made between the current and future beneficiaries. Mahalinobis matching will also be employed.

As such, I would like to ask how many more control samples should I add over the treatment samples? Also, my goal sample size is 300 to 400 farmers.

I am trying to evaluate impact of an intervention that was implemented in very poor areas (more poor people, undeserved communities). In addition, the location of these areas were such that health services were limited because of various administrative reasons. Thus, the intervention areas had two problems: (1) individuals residing in these areas were mostly poor, illiterate and belonged to undeserved communities; (2) the geographical location of the area was also contributing to their vulnerability (as people with similar profile but living elsewhere (non-intervention areas) had better access to services. I have a cross sectional data about health service utilization from both types of areas at endline. There is no baseline data available for intervention and control. I am willing to do two analyses: (1) intent to treat analysis: Here, I wish to compare the service utilization in "areas" (irrespective of whether the household in intervention area was exposed to the intervention). The aim is to see whether the intervention could bring some change at "area" (village) level. My question is: can I use Propensity Score Analysis for this? (by matching intervention "areas" with control "areas" on aggregated values of covariates obtained from survey and Census?). For example, matching intervention areas with non-intervention areas in terms of % of poor households, % of illiterate population, etc. The second analysis is to examine the treatment effect: Here I am using Propensity score analysis at individual level (comparing those who were exposed in intervention areas with matched unexposed people from non-intervention areas). Is it right way of analysing data for my objective?

Hello, I am working on propensity score-matching, regressing the pscore on several outcome....but I get diffeent ATT with Kerner al Radius...the problem is not the size of the effect, which does not have to be necessary the same, but the direction...sometimes it gives positive, sometimes it give negative. I looked at the vars that I used to match or as outcome, and also addedd the "seed"...but no changes. do you have any ideas or what could be wrong? 

Hope you can help!

Thank you

Hello everybody. I need some help with score matching. I have to calculate the p-score in Stata but have three different treatments. The idea was to do a series of binary models but I am not sure whether I am doing correctly.

I have treat1, treat2 and control. My idea is to compare treat1 with control and treat2 with control....

After that I have the propensity scores of the two binary cases what should I do? and if I use psmatch2, how do I keep the values of the pscores of the first binary model when I run the second one?

Could you please clarify the steps to follow to get the final model?

Hope you can help.

Thank you.

Stefania

which level of data agregation is better? for example if i conduct a propensity score matching, should i performed it to match students or schools?

Dear all,

i would appreciate any recommondations for good R packages doing matching on three groups (one control and two treatment conditions).

Many thanks in advance,

Manuel

Two populations, very different sizes (n = 9000, n = 300). I wish to compare the relationship between being in either population and BMI. If I control for demographics (e.g. age, race, education level) I find a difference between both populations; being in A associated with a lower BMI of 1.37 compared to being in B.

Applying PSM, I find a difference of 1.23 a.k.a. smaller. How do I explain what the added value of Propensity Score Matching is compared to controlling? 

Thanks for your expertise! 

Currently I have a sample, which is extremely unbalanced between the treated and the control. The number of treated observation is only 55, while the number of observations in unmatched control group is more than 3200. I am considering using 1-1 nerest  neighborhood matching to esimated the treatement effect. However, I am worrying that the sample structrue may impact the performance of PSM.

Is the number of treated observations too small for PSM? Does the proportion of treated group affects the performance of PSM? Intuitively, I think the large number of control observations enhances the probability of finding good matchs for the treated. Is this correct?

I have 3 treatments, saying A, B and C. I estimated the propensity score for each treatment using mlogit model and kept obsevation falling withing the common support (maximun of the minimum et minimum of the maximun of the pscores). Now, I want to test whether my psocres balance across the three groups. Does anyone has any guidance to provide? How one can do the balancing test in stata for three groups.

Thanks!

The treatment is comprised of four banks policies and outcome is risk taking behaviour of banks. How Panel data and multiple treatment should be taken care of at the same? Can we deal with it in Stata or some different software is needed. Any help will be much appreciated.